Enrollment for the next cohort of Supermanagers closes today. If you’ve been on the fence, this is the last live cohort until fall. (And if you can’t make the session timing, the build guides are designed to work on your own time too.)

I love teaching this course because we’re at a historical inflection point for knowledge work. A range of futures are on the table — some scary, some better than we can imagine — and I genuinely believe today’s managers will play the biggest role in deciding which one we get.

For all the handwringing about AI destroying jobs, I’ve seen very few practical answers to: “what would a company where AI doesn’t destroy jobs actually look like?” This surprises me, because I don’t think the answer is mysterious. I’ve worked with plenty of teams where AI enriches human-centered work, even as it transforms the shape of that work.

Some of these models are:

1. Agents that coach humans through learning a new skill

2. Deliberately designed Skills that evaluate and give feedback on human-created work, raising the quality bar across a team

3. Agent agencies: a human briefs, the agent produces, the human reviews and approves across rounds of feedback

4. Agents that create the first draft of work before handing off to humans who are responsible for the last mile

5. Or the inverse: a human sets the structure and writes the first 10% to establish the standard, then AI extrapolates from there

These produce better work than humans or robots alone. They also create a virtuous loop: the system improves the human’s work, the human’s work improves the system, and so on.

None of this is guaranteed. The default I see is wildly unimaginative people panic-slashing headcount and hoping AI gets smart enough fast enough to magically sort out the damage.

I understand the technical and economic pressures pushing toward that path. But I don’t find it convincing. Why would people (outside of those who stand to gain financially from AI proliferation) go along with this? The political backlash is already starting. Additionally, as I’ve written, there is competitive opportunity in keeping humans in the loop, even where it’s not technically necessary. Plus, humans are uncannily good at inventing new work for ourselves; we have been for centuries.

We know how the CEOs will behave. We know what the la stated goals are. The big question mark is the people in the middle: how will the managers actually tasked with galvanizing their teams around AI do it? There are hundreds of decisions buried in that work and it will be their job to figure it out.

I’m not suggesting managers are going to successfully collectively bargain with the tech billionaires. But practically, work happens in the details — and it is wild to me how few people have a point of view on the details. If you’ve ever built a product, you know it lives or dies not in the executive mandate but in the pixels. AI at work is a product, and every detail — what the agent does, what the human does, where the handoff happens, what the output looks like — is a pixel.

Today’s managers have a historical opportunity to shape what work becomes. Those who understand the technology well enough to think creatively and ambitiously about it will chart the course.

If this excites you, join us.

xoxo,

hils

P.S. Scholarships are still available if the price is a barrier. Apply here. I especially welcome applicants from nonprofits.

Happy Sunday!

Enrollment for my Maven course, How to be a Supermanager with AI, closes on Tuesday at 12 PM ET.

This is the last live cohort until the fall. (I’ll run a fully async version this summer for people who can’t do live.)

Here’s what I’ve been hearing from managers right now:

Your team started using AI. So have other teams. And…it’s causing chaos. You have more work than ever to review, and you’re overwhelmed by the obvious AI-generated workslop you’re expected to take seriously. Everyone is using AI in their own idiosyncratic way. Maybe you have a centralized AI team trying to wrangle it, maybe you’re on your own. You’re doing more cognitive labor than ever, and somehow it feels like you’re getting less leverage out of the technology than your team is.

I think this is the central management problem of the moment, and I’ve completely rebuilt the next cohort of Supermanagers around it.

Here’s what we’ll cover:

Week one: Come up for air. Build the tools and automations that claw back your time and give you some control over the treadmill of your ever-growing to-do list.

Week two: Delegate or drown. Translate delegation best practices for humans to AI: how to identify good candidates for work to hand off, how to manage context, and how to define success clearly enough that the work stays good. Plus, how and where to keep humans in the loop.

Week three: Stop the slop (!). A good manager sets a high bar for quality and holds the team to it. This week we build tools that evaluate work against that bar and coach people on how to get there. We’ll also cover what to do when people fall short, and when the work becomes “obviously AI.”

Week four: Control the chaos. Many orgs are struggling to keep their context and agents organized. This isn’t a new problem — humans have always operated from different and sometimes contradictory levels of context inside organizations — but AI makes it sharper. We’ll learn how to keep your AI efforts from descending into chaos, and how to design systems that don’t atrophy our brains.

Each week includes a two-hour live build session where we make tools you can use immediately, plus video + interactive resources between sessions.

All students get lifetime access to course materials (including anything I make for future cohorts) and a spot in the alumni Slack.

If you’d like to join but cost is the blocker, you can apply for a scholarship here.

Reach out with any questions. I hope to see you there!

xoxo,

hils

Some housekeeping:

1. Earlier this week I launched a program I am very proud of. It’s called Couch to 5K for AI and it’s the thing I have wished existed since I started working in AI. It’s a free, 30-day program that takes fewer than 10 minutes a day and gets you from “I keep meaning to figure this AI thing out” to actually changing the way you work. I wrote up my thinking about it in Lenny’s Newsletter, or you can try it here.

2. If you want to go deeper, the next cohort of my Supermanagers course starts on Wednesday. Enroll here if you want to learn how to actually use AI thoughtfully — not the hype-y version you’ll find scrolling YouTube :) If cost is an issue, I have a few spots for scholarships! Fill out this form and I’ll get back to you in a couple days.

Does your boss have AI brain?

Last week, Rachel Karten wrote a Substack post about how bosses with AI brain are making everyone miserable. These bosses are outsourcing their own thinking to AI and begging their teams to do the same; one goes so far as to put memes into ChatGPT to find out if they’re funny. Bleak!

Rachel writes:

I asked some of the marketers I spoke with if their boss’s use of AI has changed their perception of them. Many said it did. When the person who is supposedly in a role because of their experience and intelligence is constantly outsourcing their instincts to AI, it says to a team, “I don’t trust my own opinions on humor, storytelling, and marketing”.

I feel for anyone who is enduring this right now. Anecdotally, when I ask people how they feel AI efforts are going at their company, they often let out a long sigh. But it doesn’t have to be this way!

This is a solved problem. There are some very practical ways to use AI to help your team do better work — and while the consequences of getting this wrong are pretty grim (more on that later), the fixes are not complicated. I’ll walk through a few of them here.

There’s a better way, I promise

The AI-brained bosses in Karten’s piece have it exactly backwards. If you’re a leader, you presumably have expertise that’s useful to the people who work with you — and AI is a genuinely great way to scale that expertise across your team. Done well, this is doubly transformational: it saves you a ton of time, and your team gets actionable feedback on demand, as many times as they need it.

However, this is very different from simply telling people to “run things through Claude.” I cannot emphasize enough the degree to which giving people that as direction will cause them to lose respect for you.

The key is to take what’s in your head and turn it into a tool people can actually use. Until recently, the only way to do this was to write a doc or build a template — painful to make, and often a waste of time because nobody read, implemented, or updated them. You were just creating more work for people who already felt buried.

A tool is different. One thing I repeat throughout my course is that if you’re making a tool for someone, it needs to solve a problem. Importantly: their problem, not yours.

I originally explored this concept through Custom GPTs like Hilary’s Superpower Finder and Wally the Writing Partner and Slacky. They’ve been used tens of thousands of times, and they’re still just a small slice of what’s possible. (The prompts for all three are behind the paywall.)

No more docs

I keep telling people I’m “post-doc” — meaning, any time I’m tempted to write a doc to convey information to other people, I build an interactive website instead, usually with Claude Code.

When I started making prompting guides for my course, I ended up with these unwieldy beasts. Different tabs depending on which tools you used, blanks to fill in across different sections, things to copy and move around the page. A pain.

Now my resources include things like this, which updates the whole guide based on what you select:

Or this. You personalize it once, and your edits fill in across all the remaining prompts:

Or this, which lets users see the difference between prompting styles instead of taking your word for it. (This one teaches a specific move: extracting principles from good examples of a kind of work before you start on that work.)

Or this, which lets you teach concepts like problem decomposition to a room full of people who all have different problems. The prompts further down the page fill in based on what you put here.

I used AI to make all these guides, but none of the guides themselves use AI. They’re doing what I always wished a doc or template could do. And because AI helped me build them, they were faster to make than the “best practices” doc I would have written instead. They’re also actually useful, which docs usually aren’t.

I would never have thought to build any of this if I were using AI to do my thinking for me. The whole reason it works is that I’m using AI to be more imaginative, not less.

What other problems can better-packaged information solve?

One of the biggest problems I see in most companies is that very few employees have a real grasp on how the business actually works. How does the company make money? How does it spend money? What happens if you pull on pricing, or enter a new market, or shift from acquisition to retention? Where are the actual bottlenecks to growth?

Ideally, employees would understand all of this and how their own decisions fit into it. The fact that they don’t is the source of a lot of internal arguing. Is X a good idea? You can’t answer that in a vacuum, but two people will try anyway — and what looks like a disagreement about the idea is usually a disagreement about something else. One person says it’s a bad idea; they mean it’s unlikely to succeed and they’re risk-averse. The other person says it’s a good idea; they agree it’s unlikely to succeed but think the upside is worth it. They actually agree on the idea. What they disagree on is how much risk this team should be taking. And they have no shared way to even surface that, let alone resolve it.

The best attempt I have seen at solving this came from Parsa on the Corporate Strategy team at WHOOP. He built a whole slide-based WHOOP EDU curriculum that exhaustively catalogues how the business works. It’s great. I’m a huge fan.

But, as I mentioned, I am now post-doc.

So lately I’ve been building simple simulations that let people run different scenarios and see the projected impact on the business. These were some of the most helpful exercises we did in business school.

(Fun fact: my now-husband and I were on the same factory simulation team; we performed terribly because ONE OF US fell asleep on the overnight shift, I won’t say who.)

These are useful for teaching business instincts in general, and even more useful in specific situations where someone needs to make a real decision.

For example: I built Google-opoly with Lovable. I one-shotted it, I have no insider information, and the model powering it is mostly nonsense — but it illustrates the point, which is that you can build tools that help your team make better decisions much faster than you can write a doc that does the same job. (I can’t bring myself to paywall this prompt because it’s so simple. I literally just typed into Lovable: “build me a simple simulation to help me understand Google’s business and growth levers.”)

…..By the way, if you want me to build a good version of this for your company, reach out. I’d love to.

so what do you do with all of this?

Here’s the recommendation underneath all of this: use AI to make humans more valuable, not less essential.

I want to spend the rest of this essay on why I think this matters more than just "your team will be happier." Because I think the way we're collectively learning to use AI right now is going to determine a lot about what the next decade actually feels like.

The AI-brained boss move is to point your team at ChatGPT and call it a day, and I get the temptation, especially when you are underwater. But that move ends in a place none of us actually want to live. The version that serves you (and your team, and the broader project of human work mattering at all) is the opposite. You take your valuable expertise, and you find ways to extend it to your team through tools they can actually use.

Sometimes when I talk about this, people bring up the bitter lesson, the idea that attempting to “teach” AI is a losing strategy compared to throwing a bunch of compute at the problem and letting it figure out the optimal path by itself. (I like this newsletter post on the bitter lesson vs. “the garbage can” that is most enterprise organizations.)

They’re right, technologically speaking.

But like. If the people who stand to become wealthy and influential from superintelligent AI still find the lesson bitter, how do you think everyone else is going to take it?

I’m not arguing that we can somehow outsmart the bitter lesson. I’m arguing that “what is technically possible” is only one lens for looking at the future. Economics matter. Sociology matters. Theology matters. People need to see a future where they are valued. They need to see a future where their children are valued. And if the only way they see to stop it is to halt technological progress by any means necessary, well, we shouldn’t be surprised when that gets ugly.

This is why I think most of what people are being taught about AI right now is bad. The boss who tells you to “run it by ChatGPT” is one version of bad. The YouTube hype-bros who promise you’ll 10x your output by Tuesday are another. They have very different vibes, but they share an assumption: AI is the source of the value, and you are basically a button-pusher.

There’s a version of the next decade where most knowledge workers become button-pushers because nobody with influence fought for anything else. And there’s a version where human intelligence stays in the conversation. Which version we end up in depends on choices each of us is making right now.

So here is one thing you can do, right now. You probably have something you are trying to figure out, or something you wish people on your team understood. Don’t reinvent the wheel. Take any of the things I just showed you (Couch to 5K, the GPTs, the prompting guides, Google-opoly) and paste it into Claude or ChatGPT. Ask the model to help you identify the principles behind why it works, and then talk through how those principles would apply to your specific problem.

It’s time to learn like our humanity depends on it.

xoxo,
hils

Prompts for Wally the Writing Partner, Slacky, and Hilary’s Superpower Finder are behind the paywall.

Read more

The next cohort of my Maven course, How to be a Supermanager with AI, begins May 6. I just heard from a student in my previous cohort, “I’m not exaggerating -- the mindset shift has changed the trajectory of my career.” This could be you! Enroll here!

I finally got an OpenClaw (actually a Hermes Agent, but similar deal). Two things were immediately clear to me:

First, as others have said, it is helpful to think of the Claw like an employee who has an outcome they are responsible for, rather than an all-purpose assistant that you have to keep coming up with tasks for.

Second, when you delegate work to it, it fails in the two ways employees tend to fail when they do not know how to manage up: either it makes bad assumptions and runs with them, or it starts asking you a bunch of open-ended questions, which is exhausting. If you tell it, “Create and execute a marketing plan for me,” it will either run ahead with something pretty sloppy, or it will start asking you, “what channels?” “what media types?”

The problem is that your Claw does not know how to surface a decision in a way that costs you almost no cognitive effort to weigh in on.

Subscribe now

You can, of course, teach employees to manage up. Then they come to you with “I’m planning to do X for these reasons. The alternatives are Y and Z. Here’s the part I’d want your input on.”

With the humans, there is at least an excuse, which is that an entire school of management training actively teaches managers not to share their POV. The idea is that if you tell people what you think, they’ll get dependent on you, and you have to make them figure things out themselves. So instead of saying what you’d do, you ask Socratic questions. “What do you think we should do?” “What have you tried so far?”

I have a lot to say about why this is bad advice (separate post for another day) but the relevant point here is: managers who are trained this way never give their reports a model for how to bring someone into a decision efficiently. So the reports learn to either guess or punt, and then they grow up and become managers who do the same thing, and so on.

Claws are doing the exact same thing for the exact same reason. Nobody told them how you want to be brought into the loop.

So tell them.

Here is the prompt I give mine, which you could give to a Claw or whatever AI you use. It also works as a perfectly good thing to say to a new direct report.

Any time we’re working on something new, follow these steps:

1. Repeat back to me what you think I’m asking you to do, and what you think success looks like, so I can confirm we’re on the same page.

2. Once I confirm, tell me what steps you’ll take to accomplish the goal and what inputs you’ll use.

3. Break the work down into way, way smaller steps than you otherwise would. Like, 10x smaller.

4. Before each step, tell me your recommended next action, your two best alternatives, and why you’re recommending the one you are. Frame any tradeoffs in terms of what you understand I care about, and abstract out details that are unlikely to matter to me. Phrase it so I can just reply “yes” and you have what you need to move forward.

5. As we work, in a separate doc, keep track of where I accept your recommendations and where I push back, and what that tells you about me.

6. Over time, draw on those observations to check in less when you’re confident I’ll agree, and more when you think I’m likely to weigh in.

A couple things worth saying about why this works.

Step 1 exists because you are probably not being as clear as you think you are being. Asking the agent or person to repeat back to you what they heard is how you can calibrate whether they interpreted your words the way you meant them.

Step 2 reduces wasted work by making sure you are not expecting a completely different approach.

Step 3 allows you to quality check at a granular level, reducing the steady decline toward slop.

Step 4 is what allows you to maintain a high quality bar without having to think from 0 every time. Open questions, like “what tone should I use?” mean you’re starting from a blank page, which requires unnecessary cognitive work. A recommendation with alternatives — “I’d go dry and direct over warm or formal, because your past writing leans dry” — lets you reply “yes” and move on or “no” and a quick correction. Same decision but with a fraction of the energy.

Step 5 and 6 create the feedback loop that help the agent lean on you less and less as it learns how you think and work. This is helping it, essentially, develop judgement.

What I have found kind of wild about all of this is that the prompt I am giving my Claw is basically a description of what every manager has always wanted from a junior employee, and yet very few people ever receive this level of direction. The Socratic orthodoxy I mentioned earlier convinced a generation of managers that telling someone what to do would rob them of the journey, so most people never got an explicit model for how to bring a decision to a boss efficiently. The Claw has none of that baggage. You can just tell it. And once you do, you realize you probably should have been telling the humans all along.

I tell people I am a micromanager of agents and I mean it — I want to weigh in on a lot. But I am also impatient, and I never want to block an agent unnecessarily. Those two things only coexist if the agent is good at managing up: pulling me in for the decisions that matter, keeping me out of the ones that don’t, and getting better with every rep.

xoxo,
hils

More Writerbuilder

Quick housekeeping:

1. The next cohort of my four-week course, How to be a Supermanager with AI, starts on May 6. Learn to build an AI Management System that invests in your people rather than sidelining them! Enroll here.

2. We have a few tickets left for our IRL AI workshop in Brooklyn this week, Grrls in the Loop. Learn to build apps and agents that make your life easier and your work more effective. If you are self-funding, use this discounted payment link.

My former colleague Meg once recommended I read Discussing Design by Adam Connor and Aaron Irizarry, a book about how to have better conversations about design. It was so good, in fact, that I never gave the book back to her, something I feel bad about like three times a week. Sorry Meg.

The book’s premise is simple: critique is a skill. Which is funny, because a lot of people stake their whole identity on being critical despite being pretty bad at it.

Connor and Irizarry’s core move is to separate critique from reaction. Reaction is gut — “those colors are ugly,” “this feels off,” “I don’t like it.” Reaction can be informative, but it isn’t critique. Critique is analytical: it compares a design to the objectives it was trying to accomplish, and examines where the design is or isn’t hitting those objectives.

Subscribe now

Which means critique has a precondition. Before you can critique something, you have to name what the thing is trying to do. You can say: my understanding is this is trying to accomplish XYZ, and here is where it’s succeeding or falling short against that. Without a stated goal, you don’t have critique.

what is slop?

People refer to AI writing and AI design as “slop.” Originally, that word meant something specific —the output was very bad. Videos of Shrimp Jesus level bad. Over time, the models have gotten much better at making good outputs with minimal prompting. But we still call those things slop. “Slop” basically means, now, anything that is recognizable as AI — regardless of the objective quality level.

Which is weird, if you think about it.

Everyone hates AI writing, including me. It is not “good.” But is it bad? I mean, if I fell asleep in 2021 and woke up today, and someone showed me an AI-generated essay and told me a computer wrote it, I would be floored. I would think: wow, this is much better than what the vast majority of people could accomplish.

When I rolled out Slacky, my custom GPT that tightens up Slack messages, the people using it immediately started sending me clearer messages. The AI was taking their rushed, dashed-off-between-meetings writing and making it better. (For the record, one of my top-three most-used prompts remains “edit this for clarity.” AI is very good at that.)

Whenever I talk about AI writing online, I get yelled at. I’m in and around social circles of people who care deeply about the craft of writing, and they haaaaaate AI writing, and find it weird that I would defend something so aesthetically rotten. But this is where I like to uno-reverso their snobbishness back at them, and remind them that criticism in a vacuum is midwit stuff. You cannot say something is bad without understanding what it is trying to accomplish.

Yes, AI writing is stylistically atrocious, especially when you move beyond the atomic unit of the sentence. But if the goal is to convey information in the clearest way possible, then AI writing is quite good, and in fact far better than what the vast majority of people write. And for 98% of writing, clarity matters more than stylistic distinction.

Even at the sentence level, AI is drawing on effective rhetorical strategies when it defaults to “it’s not X, it’s Y.” That construction was not considered bad writing three years ago. It’s closer to objectively good than objectively bad. So it’s strange, when you think about it, that it’s now widely considered bad writing. Offensively bad writing, even.

design slop

The same thing is happening in design, though it’s more of a moving target.

When vibe coding first emerged, the purple gradient was a dead giveaway. The number of times I landed on a website, saw some purple gradient, and thought ugh, this vibe-coded-ass website, gross… Then Anthropic released their front-end design skill, and everyone was like, WOAH, design is saved!! No more purple gradients!! I can finally vibe-code a beautiful website!!

But pretty quickly everyone realized those websites all kind of looked the same, or at least had specific tells.

Last week, Anthropic released Claude Design, and the reaction was: WOAH, now I can REALLY design LIKE A REAL DESIGNER!! People started breathlessly posting their designs online, and everyone was wowed. But then they started noticing …. certain patterns … showing up in everyone’s work.

(Source)

(Source)

Notice the one italic serif word among many bold sans-serif words. The all-caps eyebrow label with a leading dot and bullet separators (NEW · MAX-CUSHION TRAINER · EMBER FLARE // MISSION CLASS · CREWED LUNAR FLYBY).

the problem with predictability

Over the weekend, I tweeted:

today’s amazing new AI-designed artifacts will look like slop in a month, once everyone learns to recognize the patterns the model falls back on. like AI-generated writing, the output isn’t objectively “bad,” (in fact it is often technically quite good), but once it becomes predictable, it reveals itself as recognizably “AI.”

this is undesirable because it exposes two separate skill issues:

1. the person lacks the design (or writing) taste to realize their work reads as obviously “AI”

2. they also lack the prompting skill to steer the model away from its default patterns

this is why there will always be a signaling arbitrage opportunity in keeping a human in the loop for creative and many kinds of knowledge work, no matter how good the tools/models get

The funny thing is, it’s actually quite difficult to prompt yourself out of this problem. You can’t say to the AI, “don’t make this design look too AI-generated,” because your definition for that is two days old and the model has no idea what you mean. You can feed it the Wikipedia page on signs of AI writing and tell it not to do any of that, and it will sort of listen.

Most people don’t even have the vocabulary to give specific enough feedback to explain what the model is doing that they don’t like. Like, what do you even say? Don’t use italics... at least, not in the way you are currently attempting to use them?

Maybe a George Costanza-inspired prompt is your best bet: “Do the opposite of whatever your first instinct is.” Make no mistakes.

the precondition for creative work

All of this makes it harder to do good creative work with AI.

A few years ago, if you wanted a design to feel sophisticated, you could Google reference images, find the obvious ones, and rip them off. But now you can’t. The obvious references are already baked into the model’s defaults, and showing up with those defaults will get you flagged as AI.

The only way out is to have a better reference set than the crowd. You have to pull the model away from the obvious reference, which means you need to know what the obvious reference is, AND you need to know what’s better than it, AND you need to be able to articulate it. Doing the work to really know the underlying craft — the history, the references, etc — is the only way to stand out.

Which means aesthetic curation has become more valuable, not less.

I really enjoyed the recent episode of How I AI with Jamey Gannon, because I’m very curious about how people get AI to look aesthetically not-garbage, and she actually shows you. She puts a huge amount of work into mood-boarding the exact vibe she’s going for. And her ability to do that mood-boarding is obviously based on years of learned knowledge about visual design.

My #1 concern about AI is that we will always have the option to “hit the easy button” to avoid doing hard work, and if we keep hitting the easy button, we as a society are going to have a very bad time. But Jamey is not hitting the easy button. The people doing genuinely good work with AI are not hitting the easy button. They are pouring tremendous work into their own skills and craft, and that investment is going to pay off more and more as things get noisier — because the ability to do actually good work is going to become rarer, and therefore more valuable.

I’ve come to think that the traits we used to call slow advantages, like being well-read, having a strong reference set, and really knowing the history of your craft, were always valuable but never urgent. They compounded quietly over years. But in a world where anyone can produce the obvious thing instantly, they become the only thing that can pull your output away from the model’s defaults, which makes them the difference between work that reads as cared-for and work that reads as whatever the computer handed you.

The people who have done the hard work of knowledge-building are not really racing against AI. They’re racing against everyone who thinks AI will let them skip it, which is a much easier race to win.

I think Diplo was getting at the same thing in a recent interview, talking about why the best DJs right now are older than him:

“I used to think there was no career. I was like, yo, I don’t want to be a DJ at 40, that’s whack. That’s a young thing. But the DJs that get older, our repertoire, our knowledge of music, is what makes us good. Then I realized, damn, being a DJ actually means having that vocabulary.

My gift is that I came from digging up old records, going house to house, passing out business cards, ‘Yo, I can clean your garage for you.’ I was waking up at 5 a.m. going to flea markets, going to Sound Library, A1 Records in New York, reading liner notes. Stealing records from libraries, looking at who’s writing on these records, understanding who David Axelrod was, who Carole King was, who the bass player on this record was, why do they sound like this, who produced this record, what label was this. I was addicted to the information.

That is what makes me good at using AI. Because I know how to prompt something specifically into a time period, a producer, a sound. Young people don’t know that because they don’t read liner notes like I did. They kind of know, ‘oh, this artist is doing this,’ but there is a skill to it. You have to have knowledge. You have to have history.”

The vocabulary Diplo spent years accumulating is exactly what lets him steer AI somewhere its defaults won’t take him. Younger DJs with access to the same tools can’t get them to do what he can, because they haven’t done the work that lets them name what they want.

show you care

I have long believed that the most important slide in any slide deck is the title slide, because it sends an important signal to your audience that unlike everyone else’s templated slides, yours is worth paying attention to.

I do not think this is baseless hypemongering or a shallow trick that comes at the cost of substance. I love this quote from Patrick Collison:

“My intuition is that more of Stripe’s success than one would think comes down to the fact that people like beautiful things, and for kind of rational reasons. What does a beautiful thing tell you? Well, it tells you that the person who made it really cared, and you can observe some superficial details, but they probably didn’t only care about those and then implement everything else in a slapdash way. And so if you care about the infrastructure being holistically good, indexing on the superficial characteristics that you can actually observe is not an irrational thing to do.”

Beautiful things signal that someone cared — and someone who cared about one thing probably cared about the things you can’t see, too. Indexing on those care signals is rational, even if it feels shallow. This is also why encouraging AI use and looking down on AI-generated content are not contradictory positions. In a noisy, low-trust world, finding ways to signal real effort is part of the work.

If you want other people to care, you need to signal that you care. Sure, the substance should speak for itself, but people are busy. They have a lot on their minds, and are being inundated with information and content at all times. Why should they stop focusing on their other priorities and focus on yours instead?

Because you care.

don’t let the slop cannons get you down

This brings us back to the people who yell at me for saying that AI writing is not bad writing. Again, I’m not sure why I’m defending it, because I too recoil in horror when I read it.

But, to once again uno-reverso my haters, if you care about writing you should care about what words mean! And BAD is not the right word here. The problem with AI writing is that without thoughtful prompting, it signals low effort. Weirdly, this is an inverse of previous low-effort signs. Typos, lack of polish used to indicate that the person writing you didn’t care that much. Now polish is suspect.

But as has always been the case, asking someone to read something that you did not put time or effort into is presumptuous. Do not do it!

I think all of this is actually good news for anyone who cares about good creative work.

Signs of low effort and signs of high effort have always existed. AI just makes them louder and easier to spot. If you’re someone who cares, you probably already show signs of high effort. The low-effort people are more visible now, sure, and it’s easy to look at them succeeding in the short term and feel exasperated.

Don’t pay them any mind. They are playing a different game than you are, and letting them stress you out will only undermine your ability to play your game well.

You are not in the race against slop cannons. Care and effort are more valuable by the day.

xoxo,
hils

PS — Meg, if you’re reading this, I still have your book. I really am sorry.

More writerbuilder

A few quick notes:

1. If you want to attend Grrls in the Loop but can’t get it covered by your company, we now offer a self-funding payment option here. If you’re still not able to make this work with your budget, reply to this email and we’ll figure something out.

2. I’ve also opened a new cohort for my Supermanagers course starting May 6. Enroll here.

3. I’m doing a free lightning lesson on How to Self-Promote Without Burning Trust with Ben Erez and Mallory Contois. Sign up here.

Ahh.... flow state. Is there anything better? I love being in flow, where I get totally lost in something I’m working on and completely lose track of time. I have found various ways to tap into flow state in my life; writing is the most reliable one, but I’ve had a variety of crafty phases, including illustrating and beading. Unfortunately most of my flow state onroads involve me sitting quietly in a chair for hours on end, but presumably you could find flow state while exercising, too. I just love the satisfaction of sustained, uninterrupted focus. It feels so rare in this day and age.

Losing flow state was a primary fear I held about motherhood. Babies, cute as they are, interrupt. Creativity requires flow. Therefore, it felt like in order to gain access to a hugely soul-affirming part of life (kids) I would have to give up another hugely soul-affirming part of life (creativity).

With apologies for this jarring pivot, this is why I am awed by AI. The technology has allowed me to bend the shape of my time and attention. I used to feel like my brain was a finite resource, and so focus was a zero-sum game. To give attention to something, it had to come at the expense of something else.

Subscribe now

And I still feel that way -- in fact I think focus is more important than ever -- but I do not think about the tradeoffs in quite the same way. It feels more like I can pick and choose what takes up space in my brain; the most precious stuff: my family, and conversations with friends, and work and hobbies that make me light up. And the other stuff, well, I have better options now. My other brains, my little robot friends, can take care of those.

Not in an “outsourcing your thinking” way. I hear so much concern about outsourcing your thinking, but that is not my experience of using these tools. It feels more like the way photos augment your memory. My strongest and most vivid memories are those where a camera was present. Sometimes I have no memory at all of an entire vacation I took as a child, until I stumble on a photo album with pictures from it, and then suddenly all these memories come back. The technology is preventing an unnecessary loss. Maybe it comes at a cost, I don’t know. I do wish everyone didn’t record concerts on their phones. But I am glad if one person records the concert. I’m happier with the memories.

Anyway. I digress. Today I am here to talk about Virginia Woolf.

In A Room of One’s Own, Woolf argued that women needed £500 a year and a room with a lock on the door to make art. Economic independence, and physical space. She illustrated this by imagining what would have happened if Shakespeare had a sister with equal genius. That sister would have been denied education, forced into domestic duties or marriage, mocked for her ambitions. Even if she’d somehow managed to write, she would have been interrupted constantly -- by children, by household duties, by the expectation that her time belonged to everyone but herself.

The argument was radical because it was so unflinchingly material. Woolf didn’t say women needed to be braver or more inspired. She said they needed money and space. Without economic independence, women couldn’t afford the time to write. They had to work, or they had to rely on men who controlled their schedules. And without a room they could lock, they couldn’t protect their attention from the constant demands of domestic life.

The point wasn’t metaphorical. Woolf was saying: you cannot make art in stolen moments between other people’s demands. You need sustained, uninterrupted time. You need a door you can close. You need the material resources to protect your attention from the world.

A century later, we’ve made real progress. More women have money, education, rooms. However. You can close the door, but modern life is intrusive. Your phone is still pinging with the endless list of things you’re on the hook for. Your brain is still tracking the mental load of what needs to happen after this stolen hour ends.

Women still carry the fragmented attention of domestic life. Even when we’re not literally interrupted, we’re interruptible. Our focus exists in the margins between other people’s needs. We work in the cracks.

This is where the technology comes in.

AI tools don’t require your continuous presence. You can direct them in 30-second voice bursts while doing dishes. You can set them running in the background while you’re doing something else entirely. They accommodate interruption instead of punishing it.

I dictated sections of this essay while holding my baby. I asked an AI agent to pull together research on Woolf’s economic arguments while I was feeding him, and another to fact-check it while I rolled around on the floor with him. I sat down this morning with one hour to write, but instead of starting from zero, I had a (bad) rough draft and a bunch of targeted research. I rewrote the entire essay myself (again, I really love writing, so the idea of having a robot do it for me is unappealing) but I was not starting from a blank page. There is no way I could have written this essay from scratch in one hour. But I could do it with help. The AI is not my ghostwriter, it’s my film crew.

I am not suggesting we should happily accept a post-focus world. Sustained concentration is valuable and I think most people are too quick to let their aptitude for it lapse. But access to long stretches of uninterrupted time has always been unevenly distributed, and that unevenness has always been gendered. Woolf knew this. What these tools change is what you can get done before you sit down. The research, the rough draft, the life admin cluttering your head. So that when you do get your one hour, you can really, actually, focus.

As I demo on How I AI, it also helps a ton with the mental load. I have a system to capture and manage all the little tasks that are otherwise floating around in my head and taking up space that could be going to more important things. It takes life admin, breaks it down into smaller tasks, and slots those tasks on my calendar so I actually get them done. It helps me avoid dropping balls, without having to constantly have a voice in the back of my head going, “hey, are you dropping any balls right now?? are you sure???”

Now, I am not recommending you lock yourself in a room and put a robot in charge of the childcare. To be clear. To be clear! But what I have found is that these tools let me get the most out of the time I do have.

Woolf’s insight was that creativity is not a personality trait; it is a material condition. The reason women weren’t writing great novels wasn’t that they lacked talent or ambition. It was that they lacked the resources to protect their attention long enough to do the work.

I think we are living through a genuine expansion of those resources.

I don’t know what Woolf would have made of all this. She was deeply critical of women who sought empowerment by breaking into spaces dominated by men, only to replicate the same structures that excluded them in the first place. So I don’t think she would love the tech industry as a whole, to put it mildly. But I think she would have understood what it means to finally have the infrastructure to match the ideas.

When it was possible to close a door and keep the world out, women needed rooms of their own. Now we need agents.

xoxo,

hils

More Writerbuilder

New: You can now grab a single-day ticket to Grrls in the Loop (Brooklyn, April 22-23). Day 1 is all about building real tools with AI. Day 2 is agents and automation. Pick the day that fits or come for both. Grab your ticket here.

In the past few weeks, there has been an uptick in professional writers getting caught using AI and summarily disgraced. I have been tracking the discourse with interest. Over on X, people wondered why some groups of workers have been quick to embrace AI while others scorn it. Professional writers do seem to see AI as evil. Compare them to software developers, who have mostly accepted or even embraced the technology. What gives?

You might think it is because these are categorically different kinds of people. But I am here to tell you they are not!

In fact, writers and software developers are the same person. What, you ask? Writers, with their metaphors, and developers, with their…math? I have worked with many of both, and I promise you they are fun-house-mirror-images of each other. Writers are Wario software developers.

Subscribe now

Both of their jobs are to write, though in different languages. As a result they tend to have strong opinions about syntax, style, structure. They can be, dare I say, precious about elements of their work that seem pedantic to outsiders. They have low tolerance for corporate politics. If you find yourself in a job where you are responsible, in some way, for the output of a writer/developer, you might find that their concerns verge on tedious. (That is not an insult; someone else’s tolerance for tedium has saved my butt on more than one occasion.)

They are the same person. The difference is that writing culture trains you to fuse your identity with your output, and engineering culture actively trains that out of you.

The warp

A software developer writes code, runs it, and finds out in seconds whether it works. Then they write a test. Then they ship it, and find out if people actually use it. The whole craft is built on a chain of increasingly consequential feedback loops, from milliseconds to months. Developers are trained by their working environment to treat their first attempt as raw material and tweak based on what they learn from the various feedback loops.

Writers have almost none of this, and the nature of the work makes it nearly impossible to build. A piece of software, if it’s good, keeps running. People use it today, tomorrow, next month. You can measure it over time. A piece of writing gets read once and the reader moves on. Maybe you get some analytics back in the form of “views” or “impressions” but it’s not totally clear if those are even real, or whether they were driven by your writing versus some downstream algorithm fluke. The quality signal, such as it is, comes from the writing itself. The primary feedback loop becomes “does this feel right to me,” and as a result the act of writing becomes inseparable from the quality of the output.

If you don’t have reliable external systems to tell you whether your work is good, you develop an internal one. That internal system becomes sacred. Your identity as a writer — your sense of self, your value in the world — gets wired into the act of producing the words.

Engineering culture goes the other direction. Your code gets reviewed, rewritten, handed off, refactored by other people. You jump in to pick up someone else’s half-finished work when they start falling behind. The entire culture is built around evaluating work separately from the person who produced it.

Making concessions

My husband was listening to a podcast recently where a group of writers were upset that Claude displays the word “thinking” while it processes a response. The idea that this machine is thinking, they suggested, is a lie, or at best a gross oversimplification.

Sure, maybe. But the people who built Claude are trying to keep users engaged while the model processes a response. They’re not scoring points for technical accuracy in a loading screen. That’s a concession writers are less likely to make. When your quality system is internal, you don’t make concessions easily, especially when those concessions require a deviation from what you believe to be correct.

AI is, of course, quite a concession.

I get it. I too find AI-generated writing off-putting. I too find it annoying when someone suggests that the thing they vibe coded in 20 seconds is as good as the work a real craftsperson spent their career learning to do, whether that’s a website, product, or piece of writing.

The thing writers are protecting when they reject AI is real. But I think the most useful thing a writer could borrow from the engineering side of the fun-house mirror is not the technology but the idea that external feedback loops don’t have to replace the internal one.

Work with your stubbornness, not against it

I am not suggesting that writers go full-on optimization machine, hurling your words into the great Plinko machine of the internet searching above all for clicks. Your stubbornness as a writer is a strength. It is what makes the work worth reading.

But. I do not think it is wise to ignore AI entirely, even if you see it as the enemy and its existence as a threat to your identity. So here are some ways to use it that can make you a better writer. All of them involve getting mad at it.

1. The spite draft. Ask AI to write your piece. Read it. Get mad at how bad it is. Write it yourself, better, fueled by righteous indignation.

2. The slop bomb. Struggling with procrastination? Ask the AI to write it for you and tell it if you do not share your own draft within the hour then it should send the AI-generated slop draft to your editor claiming it is your own work. The threat of humiliation will compel you to get over your writers’ block.

3. The internal critic validator. Start writing your sentence, then stop halfway through and ask the AI to finish it. If it finishes the sentence the way you were going to, it turns out what you were about to say was not that unique and you might need to think some more.

4. The “old man yells at cloud.” If you’re blocked, open Claude and just start talking until you figure out what you actually think. Then close the window and ignore whatever Claude said. You just needed a wall to throw the ball against. Engineers call this rubber ducking. See? Wario!

None of these requires you to publish a single AI-generated word.

Godspeed, Wario!

xoxo,
hils

More Writerbuilder

Early bird pricing for Grrls in the Loop ends Monday. It’s your last chance to save $500 on two days of hands-on AI learning in Brooklyn, April 22-23. Get your ticket today, and get tips for getting your company to cover the expense here.

Dear Writerbuilder,

My company doesn’t let me use Claude Code or Cowork or anything like that. I keep hearing people talk about how transformative these tools are, but I can’t figure out how that applies to me when I can’t even use them for my actual job. Is it worth investing time in learning this stuff, or should I just wait until my company catches up?

- Blocked

Dear Blocked,

Let’s talk about leverage.

The answer to your question is yes, it’s worth it. Having a coding agent (I consider Claude Cowork a coding agent) completely changes the shape of work in a way that feels a little abstract, especially if you haven’t spent much time thinking about code before. This is equally true for people working in finance and marketing, for engineers, for consultants, for stay-at-home moms.

Access to a coding agent is like having a team that exists entirely to solve your problems for you. The big problem that most people have is that there is too much to do, and not enough time to do it. Assuming that some of the things you have to do are computer-based, your little robot team can help solve that.

When people think about “machines helping with work,” their minds often go straight to automation. Like, Rosie the robot from the Jetsons making dinner and cleaning up the house. Imagine the thing the human does, and then imagine the robot doing it. So when they imagine a robot helping them with work, they imagine the robot doing their tasks for them.

And this is a fine place to start. But tinker around a bit, and you’ll find that when you take this approach, you have to do a fair amount of babysitting and correcting. You will ask the AI to build you a financial model, and it will do an OK job, but you’ll have to go through and make sure it’s actually good, which in many cases takes just as much time as doing it yourself.

Subscribe now

why managing AI is like managing a team

This will be a familiar struggle to anyone who has managed a team before. It takes work to learn how to delegate well, how to communicate the information that people need to make good decisions, how to set a high quality bar and empower people to meet it. So why is it worth the effort, when the alternative -- simply doing the work yourself -- is in many ways easier?

It is worth the effort because the upfront investment gives you leverage, meaning a small amount of your time and energy produces a disproportionately large amount of output. Instead of a 1:1 relationship between effort and results, you get a 1:many. One hour of good, focused work fans out across your team without requiring more of you. And the better you get at directing, the wider that ratio becomes.

The same is true with coding agents. Everything on a computer is code, which means that anything you do on a computer can be accomplished by a coding agent.

(This doesn’t mean the coding agents are building apps for you, necessarily. You might think writing code = building apps; you would be forgiven for thinking this, because in recent years that’s what we used code for, because it was expensive to write so in many cases was only worth writing if there was a product to be packaged up and sold on the other end. But this is not inherently true of writing code, and now that it’s cheap to write, everything has changed.)

from tasks to systems

If coding agents can do anything a computer can do, that means they can do your computer tasks. But -- you wouldn’t go through the effort of hiring and training someone just to have them do one rote task on repeat. That does not give you leverage. You hire them because, over time, they start to own outcomes. You don’t have to keep coming up with things for them to do. They start figuring out what needs to get done to achieve some objective, and try different approaches, learn what works, and get better at achieving that objective.

This is what coding agents can do for you. You get a return from them not when they can do tasks for you but when they can accomplish goals for you. Less “summarize this report for me” and more “monitor our competitors, flag any changes that matter, propose a response with supporting data, run your recommendations by me, and then execute.”

The latter is a system you have designed, and it is entirely possible to accomplish with where AI is today. Focusing on the specifics of tools you can or can’t use at this moment in time is a bit like asking, “if I can’t hire Hilary’s team right now, is there a point in learning how to run a team?”

Of course there is.

Automation is a 1:1 concept: one task in, one task done. That’s a fine place to start, but on its own it’s not worth the hassle. The people who are getting real value from coding agents have made the same transition that separates a great individual contributor from a great manager. They are thinking beyond their to-do list: what is the system that I can design to ensure great work gets done every day, whether I am paying attention or not? That’s leverage.

Concretely, here’s what that looks like: a marketer builds a system that monitors competitor messaging daily, flags when their positioning starts to overlap with hers, tracks which of her own campaigns are still working and which aren’t, and drafts new differentiation angles with landing page mockups ready for her to review.

why start now

Much like managing a team, it takes time and practice to do this well. So if you buy that this is what work will look like in the future, now is the time to start learning.

And, you don’t need your company’s permission to begin. You can build agents and tools that make your work better without ever touching company IP. For example, an agent that reads every earnings call transcript in your sector to tell you when a competitor is shifting strategy and whether that warrants a response. Or, a training tool that gives your team practice reps on how to communicate in a way that leads with the so-what. You can show these to people you work with to build the case for getting formal access.

Finally, being the person at your company who knows how to build with these tools across domains is one of the fastest ways to get in front of leadership, expand your scope, and make yourself impossible to ignore. (FWIW, I recently told a room of HBS students that if I were in their shoes, I would cold pitch CEOs on being their personal AI chief of staff. The window for this kind of move is wide open.)

The way we work is changing, but the future is not fixed. The people who learn to build with these tools now are the ones deciding what the future of work will actually look like. Pull up a chair.

We teach these skills at Grrls in the Loop. Come hang out with curious and accomplished women interested in building with AI. Early bird pricing ends Monday, so it’s your last chance to save $500. Get your ticket. Tips for getting your company to cover the expense here.

xoxo,

hils

More Writerbuilder

6 more days to save $500 on tickets for Grrls in the Loop, our in-person AI workshop in Brooklyn. Early bird pricing ends April 6. Here’s a guide for getting your company to cover the cost of the ticket.

Every time management system I’ve ever abandoned failed for the same reason: it was designed for the person I wanted to be, not the person I actually am. I’d set it up with the best of intentions, full of 60-minute walks and three daily priorities and neatly categorized to-do lists, and then real life would happen and the whole thing would collapse.

I was back on Claire Vo’s podcast How I AI this week, and this time I showed her something I’ve never really walked through publicly before: the way I use Claude Code to manage my life out of the terminal of my computer.

Subscribe now

What makes this setup different is that I never designed it at all. I let Claude watch how I actually behave, so the system that emerged works with me rather than against me.

The episode is a full walkthrough if you want to see it in action, and I wanted to go a little deeper on some ideas that came up during it. You can also see the Figma file with screenshots from all the demos I share (plus some extra ones that we didn’t get to) at writerbuilder.com/howiai.

1. the 10x impact test

People often ask me how I decide what to use AI for. I have a simple heuristic: if I got 10x better at this thing, would it have a 10x impact on my life? If it would not, then I’m happy to let AI handle it. If it would, that’s where I want to spend my time, so I am careful about how I use AI.

This framework also works within a single task. Giving a talk is a 10x activity for me, but not every part of giving a talk is. The ideas, the narrative, and how I connect with an audience are the parts where more effort actually makes the talk better. Formatting slides is not. So even within something I care deeply about, I’m trying to spend as much time as possible on the parts that most benefit from my attention.

Where you are on the learning curve matters here, too. I spent years formatting slides, and I don’t regret it, because at the time those reps were genuinely making me better. But at some point, for certain tasks, more practice stops producing better results. That’s when it makes sense to delegate to AI. This is especially worth keeping in mind if you manage people, because your team is probably earlier on that curve than you are, which means the work you’d automate might still be valuable learning for them.

In the podcast, I walk through how I use AI to structure my day. Here’s how the 10x test applies:

1. Track and organize my to-do list: Delegate.

2. Decide what to prioritize today: Do this myself.

3. Break big tasks into smaller first steps: Delegate.

4. Build my schedule around my constraints: Delegate.

5. Add events to my calendar: Delegate.

6. Reflect on how I’m using my time: Do this myself (though I collaborate with AI, which can spot patterns I’m missing).

This will look different for different people; tasks that are low leverage in one type of work may be high leverage in another. But the framework still holds.

(When I showed my husband the bear illustration accompanying this post, he asked, “if you got 10x better at chilling in a cool stream, would that have a 10x impact on your life?” to which my answer, to be clear, is an emphatic yes.)

2. complexity has to earn its keep

When people see a workflow like the one I demo in the podcast, they assume I spent a weekend wiring everything together. I did not. I actively resist that, because about 80% of the workflows I think will be useful when I first imagine them don’t pan out. That’s a lot of wasted effort if you build the “real” version of each one.

So my rule is: start manual and disconnected, and only invest in making it good once it’s earned its keep.

For example, you may be wondering, “how does Claude actually know what you’re doing all day?” I just tell it. With my mouth. I have Claude open in my terminal on the right side of my screen, and I narrate what I’m doing as I go. ‘Okay, sent the email. Heading off to make my return now.’ Or, when I’m wrapping up my day, I’ll check in: ‘I got everything done besides calling the doctor, so let’s make sure I do that first thing tomorrow.’ Claude hears it, logs it, and now it knows what happened without a single integration. I call this the Yapper’s API.

It’s not perfect, but it’s remarkably simple, and most of the time it gets the job done.

I only invest in real complexity after I’ve been using the hacked-together version for at least a week, I know I’ll keep using it, and I feel like it genuinely needs an upgrade. For example, I do now have my Google Calendar connected, because I was using the ‘plan my day’ workflow most days and it had clearly earned it.

I think this is why a lot of people’s AI productivity setups don’t stick. They start by building the system instead of starting by solving the problem.

3. let the AI observe you instead of configuring it

When I set up Claude to help me plan my days, I never sat down and wrote out my preferences. I didn’t create a document that said “I’m most productive in the morning” or “I need to pump every three hours” or “don’t schedule deep work after 9pm.”

Instead, Claude figured that out over time. Sometimes I’d notice something and mention it in passing: ‘Yeah, I stayed up too late last night because I got into a flow on deep work.’ And Claude would come back with, ‘That’s 4 times this month. You’ve said keeping a consistent bedtime is important to you. Want me to add a note to your preferences to avoid scheduling deep work late?’

I prefer this, not only because I am lazy, but because my aspirational preferences are different from my actual behavior. At one point, I told Claude to schedule a 60-minute walk every day. It did, but eventually it noticed I wasn’t actually going on these walks. When that happens too many times, I start ignoring my calendar entirely, which defeats the purpose. So now walks only get scheduled on days where I’m genuinely committed and other work won’t crowd them out.

I think this is the hardest thing to explain to people about how I use AI for this. I didn’t set up a brilliant system. I let the system emerge over time from how I actually behave. And because Claude is doing the observing and adjusting, the cost of maintaining it is basically zero. I don’t have to update my preferences when my schedule changes, or when my baby starts napping at different times, or when I realize that my three-priorities-a-day framework was aspirational nonsense.

The AI just notices, we check in, and it adapts.

None of this required any specialized knowledge. I didn’t write code. I didn’t build an app. I didn’t even set up a particularly organized file system. I just started telling Claude my problems, and we figured it out together.

If you want to try any of this, you don’t need a complex setup. You need a terminal, a folder, and a willingness to talk to your computer like it’s a person (Yes, I use Wispr Flow for this).

Start with one problem. And then just... tell Claude about it.

xoxo,

hils

Links:

1. How I AI: My Claude Code tutorial

2. How to set up everything in the demo

3. Figma file with all the screenshots from the podcast

4. Tickets for Grrls in the Loop, April 22-23 in Brooklyn

Join Hilary & friends in Brooklyn April 22-23 for a 2-day hands-on AI workshop for women!

My realtors are saints. We’ve been working with them for more than a year now; not only have we not bought a house, but every ~3 months or so I will text them a listing and ask, “hey, what do you think about this place?” and they will respond, “this is kind of the opposite of everything you said you wanted?”, and I’ll respond something like, “I know, ignore what I’ve said, what’s the vibe like here?”

I don’t find house hunting to be a particularly enjoyable experience, for reasons including: I have no idea what I want. Part of this is because I’m a new mom and so my housing preferences are very different from just a few months ago (shout out to everyone who advised me not to rush into buying a house while 8 months pregnant), but part of this is because the way I make decisions is very misaligned with the way real estate information is structured online.

By that I mean that I often make important decisions based on vibes. This doesn’t mean the decisions are random, or thoughtless, it just means that I am looking for a good vibe and I know it when I see it and I cannot easily discern the vibe of houses from real estate listings.

I can set email alerts for houses that meet my criteria on price, bedrooms, bathrooms, square footage. But not, “is there a cute store nearby that I can walk to when I feel like puttering around?” or “is the park nearby the kind of park that inspires local residents to make merch for it?” or “will I like the people I meet if I live here?” Even the bedroom count fails to do the job I need it to do. Are all those bedrooms pleasant, or is one in a basement that is always way too cold? Would they make for suitable offices?

The shape of problems used to be whatever shape you found them in. But now my good friend Claude Code allows me to bend problems to fit my disposition.

I’ve since made two Skills that help me with my real estate problem.

The first is my party game. The towns around Boston have very strong personalities, but as someone who moved here recently, I don’t have a good grasp on exactly what these personalities are. So, I thought it would be fun to play a little game, where Claude presents me with two hypothetical people I might meet at a party, I tell it which one I would rather talk to, and then after doing this 10 times it recommends the town where it thinks I will be most likely to meet the kinds of people I’d want to talk to at a party. You can try it with this prompt (and presumably can sub in any geographic location):

I am trying to understand which 3 towns/suburbs around the Boston area I would be happiest in. For the purposes of this exercise, assume I want to make this decision solely based on where I would meet the kind of people I would be excited to spend time with, so do not factor in proximity to the city, commute, prices, or anything else. I want you to describe two people to me in extremely specific detail, and then ask me which one I would rather talk to at a party. (Me saying “neither” is acceptable.) Do this 9 more times. After each round, update your understanding of who I like hanging out with, then create the two new characters based on the attributes you need to better understand in order to make an informed recommendation. these people should be thoughtful character studies based on the kinds of people who tend to live in certain areas, interested in the types of topics those people would be interested in, with the types of values you’d expect them to have. at the end of your questions, recommend the three towns you think I would be best suited for.

This is both ridiculous and genuinely useful. “Would I want to talk to this person at a party?” is a question my brain can answer instantly. “Which of these six towns with slightly different median household incomes and school rankings would make my family happiest?” is a question my brain cannot easily answer.

This is bending information to the shape your brain can work with. The information is imperfect, of course; I will not defend my methodology, fraught as it is with bias. But imperfect information that nonetheless paints a picture can be more helpful than factually correct information that does not.

Even more useful: I created a skill where I drop in a link to a real estate listing and Claude outputs a briefing that tells me what my life would look like if I lived there. It starts with a verdict: green, yellow, or red. And then it lays out its rationale, based on what it knows about me: Are there interesting places to walk to? What are they? How far are they? Is the street busy? Do I have to drive for groceries/fitness? What are my grocery/fitness options? Is the house a flip? Are the finishes soulless or do they have character? How might we use the rooms? What’s the vibe of the neighborhood? Are there any red flags that might not be obvious to me?

(The prompt for my /listing-brief skill is behind the paywall.)

The briefs are genuinely fun to read, and way more informative than a listing. But it also blows open the aperture of my search. Normally, house hunting works top-down: you pick a town, then you look at houses in that town. But I don’t have a town. I have a vibe. And the right vibe could show up in any of 20 towns. Without this tool, that would mean reviewing 20-30 new listings a day, most of which would be obviously wrong for us. Now I can cast a net across all of them and only pay attention to the ones that come back green.

Data helps me make plenty of decisions, but in the case of buying a house, I’m much more drawn to narratives. So I took a data problem (x BR y BA z SQ FT) and turned it into a narrative problem. Now I can make progress! And I’m not trying to replace my realtors, who again, are saints (can you imagine how unhinged my texts to them are...). I’m just trying to make informed decisions that are not constrained by the information that is readily available to me. (And, to be honest, to make a needlessly painful process a bit more fun).

I share this not to tell you all way too much about my house hunting woes, but to illustrate the creative ways AI tools and especially coding agents can change the entire shape of the problems in your work and life. I’ve been getting some questions ahead of our Grrls in the Loop IRL workshop about whether it makes sense to come if, for example, your company won’t currently let you use Claude Code, or if you’re not in a technical role.

While there are practical reasons to learn these tools regardless of your current work setup (your staying competitive for future opportunities, etc.), I think the bigger reason is that it fundamentally changes how you see problems. Learning to use coding agents teaches you to look at any problem in your life and ask, does this feel unsolvable because it’s actually unsolvable, or is the problem just coming at me in the wrong shape? And if it’s the latter: can I reshape it?

It can take time to learn how to think this way. But once you do, it’s hard to stop.

xoxo,
hils

more writerbuilder

Behind the paywall:

• The prompt for my /listing-brief skill

• How to set it up & customize it for your preferences

Read more

Illustrations and logo by Adam Eastman

Last August, some friends and I ran an in-person AI workshop for women called Grrls in the Loop. We had an absolute blast and couldn’t wait to do it again.

So we’re doing it again! April 22-23 in Brooklyn.

Based on the overwhelmingly positive feedback, we’re making it two days this time. It’s going to be so chic and fun. Come join us to meet creative and curious women from all sorts of backgrounds who want to get up to speed on everything that’s going on in AI.

Check out our website for more info about the event, including how to get the ticket cost covered by work.

Spots are limited and will sell out. You can save $500 if you get it by April 6, so act quickly!

Some of my favorite testimonials from last time:

“I feel as if today rewired my brain. I’m inspired to do many things, including passing on this knowledge with my team.”

“This was the most powerful upskilling training I’ve ever experienced. Honestly, it was worth far more than the price.”

"I walked away feeling inspired and excited about the possibilities. The training sparked so many ideas for practical use cases, not just for my role, but for teams and functions across the org."

"My mind is still blown that I created a functional prototype. I was intimidated to get started, but having them step us through it made it so much more approachable. I can't wait to keep building."

"Holy Moly, I've been playing on hard mode by NOT learning about these technologies."

97% of attendees said they would recommend it to a friend.

Get your ticket today.

Here’s what we’ll be covering:

Day 1: From Chatting to Building. You’ll learn how to identify which problems AI actually solves well, write a spec before you build anything, and then build a real working tool from scratch.

Day 2: From Building to Orchestrating. You’ll learn what agents actually are (live demo included), give your agent a job description, pick a project, and build it in Claude Code. You leave with something that runs autonomously: a real workflow that helps you with your job.

You’ll have a team of excellent coaches on hand to help you work through everything. We also added a fireside chat and a closing Q&A panel on what AI actually changes about your career and how you lead.

No coding or technical experience needed. You just show up with a laptop and a dream.

Our first event sold out. This one will too. If you’re interested, don’t wait on this.

xoxo,

hils

The job market is an attention economy now. There are so many plausibly qualified people for any given role that the hard part is no longer being good enough, it’s getting noticed. Whether you want to go work for a company or do your own thing, the problem you need to solve is capturing the attention of the people who you want to hire or pay you.

That means you have to start thinking about the entire path someone takes from never having heard of you, to learning how you think, to deciding to work with you. Like it or not, your career is a funnel now.

I can hear you groaning!! I’m drowning in a sea of groans! “Ughhh Hilary I know where this is going. Please don’t tell me I need to work on my personal brand. I hate self-promotion, and the idea of contriving a corporate-friendly personal brand makes me want to die.”

Don’t worry. I wouldn’t do that to you. In fact I am here to save you from the hordes of people telling you to look within your soul until your personal brand emerges fully formed, clad in armor, like Athena from the forehead of Zeus.

You do, however, need a strategy for how you will break through the noise and get noticed in an oversaturated job market. My humble suggestion: develop a talk.

Here’s what I do not like about Personal Brand Discourse (is that a thing?). Much of the advice you get about Branding Yourself, especially as a woman, tells you to look inward. Think about who you are and what you stand for and what you want to convey to the world. Sit with your journal and figure out your unique value proposition. Introspect!!

I think this is very bad advice, for a few reasons.

First, for this whole exercise to be successful, you need to get information from the world. You don’t know yet which parts of your knowledge have real demand, nor how to talk about them in the way that will capture that demand. You don’t know whether the things you think are interesting are interesting to others. You need signal, and you can only get signal by putting yourself out there and seeing what happens.

Second, tying your professional positioning to your identity is a trap. Because then when it gets rejected, it feels like an attack on you. You actually want it to get rejected, because that process of putting it out there, getting feedback, and iterating is how you develop it into the strongest version of itself. You can’t do that if every “no” feels like a referendum on you as a person. You will instead avoid putting yourself out there in order to protect your ego.

So here’s my alternative: instead of thinking “who am I, and what do I want the world to see me as?” think about this: “if I had to give a talk tomorrow, what would I give a talk on?

Then go develop that talk. And give it.

(Yes, I made a Custom GPT to help you figure out what to talk about.)

your mini talk circuit

Start where you are. If you’re in a job, give the talk at your job. Pick something you know well, something you’ve figured out that other people are still struggling with. “How we fixed our onboarding flow.” “What I learned managing a team through a reorg.” “How I use AI to do X.” It doesn’t need to be grand. It needs to be specific and based on something you’ve actually done.

Give this talk to your team. See what happens. Do people show up? Is there energy in the room? Do you hear good things afterward? Those are all good signs, obviously. If you get these green flags, can you give it to another team at your company? Do the green flags continue?

Then expand outward. Reach out to a friend at another company and say, “I’ve been working on a talk about X. Would your team find it useful?” This is a much easier door to open than “hey, are you hiring?” You’re offering something, not asking for something.

Over time, you refine it. You take feedback. You notice which parts land and which parts lose the room. You build it out. Eventually you develop a point of view that gets you in rooms, because you’re the person who gives that talk. And because you have practiced and workshopped this talk with friendly audiences, you crush it. So those people remember you. They mention you when a role opens up, or when someone asks “do you know anyone who’s good at X?” You’re building the top and middle of your funnel without ever having to call it that or even really think about it.

why this is better than a “personal brand”

The talk is a product, not an identity. You can test it, iterate on it, and improve it based on real feedback from real people. If you give a talk and nobody shows up, that’s useful information: there probably isn’t demand for that topic, and you should try a different one. Or maybe you just need a better framing for it. Either way, that’s so much healthier than basing your whole professional identity on a positioning that you workshopped in your journal and never validated.

It’s also lower-stakes in a way that matters. When you apply for a job and get rejected, it’s hard not to take it personally. When you offer to give a talk and the response is lukewarm, it’s just an idea that didn’t land. (I have, like, a 20% hit rate on my ideas landing. Maybe 10%.). You try a different angle or a different audience. The emotional distance between “they didn’t want my talk” and “they didn’t want me“ is significant, and it lets you actually learn from the feedback instead of spiraling.

There’s also no treadmill. When you decide to start a newsletter or a podcast or post more regularly on social media, you’re signing up for a recurring commitment. You have to keep producing, consistently, or it stops working. A talk is a single thing you develop once and give over and over. You’re not committing to an output schedule. You’re building one asset that gets better each time you use it.

And! Best of all! Whatever your talk is about, that’s probably a skill people will pay for. If your talk is “here’s how to fix your onboarding,” that’s consulting work. If it’s “here’s how I use AI for X,” that’s training. If it’s something people consistently want to hear about, that’s a signal that you’ve found a real problem, and you’re the person who knows how to solve it. The talk is how you figure out what that is. It could become something you sell as a consultant, something you write about if you want to become a content creator, or exactly what makes you stand out for your next job.

The Personal Brand Industrial Complex gets the diagnosis right: you do need to be known for something. But the prescription is backwards. Your positioning is not a truth buried inside you waiting to be discovered. It’s a hypothesis that needs to be tested. The talk is how you test it. You put an idea in front of people, watch what happens, and adjust. Over time, what you’re “known for” is something you earned in a room, not something you decided in a journal.

But it starts with the talk.

And look, I know this feels like one more thing on the pile. But most people are spending 100% of their professional energy on work that’s only visible inside their company. The irony is that the busyness that prevents you from building any visibility outside your team is the same busyness that makes you more dependent on your employer and more vulnerable if the job goes away. Developing a talk isn’t adding to your plate. It’s reallocating some of your effort from work that only your team sees to work that compounds.

why this matters right now

We often talk about “the attention economy” as it pertains to social media companies competing to attract and monetize our attention as consumers. But the economist Herbert Simon described attention economy dynamics all the way back in 1971, noting: “A wealth of information creates a poverty of attention.”

His insight was that when there are only a few options, the hard part is finding a good one. When there are thousands, the hard part is choosing; choosing requires attention, which is finite. So attention becomes the scarce resource.

This is what happened in consumer markets. Think about buying a mattress in 1995 versus today. In 1995, there were 15 options at the store. You would go in, sit on a few, pick one, leave. Now there are thousands online, and you can’t possibly evaluate them all, so you default to the one you’ve heard of, or the one a friend recommended. The scarce resource is no longer good mattresses. It’s your attention.

The same shift happened in hiring. Five years ago, applying for a job required real effort, which kept the volume manageable. Now LinkedIn lets you apply in two clicks, AI generates resumes in seconds, and hiring managers are buried under hundreds of plausibly qualified candidates they can’t meaningfully evaluate. So they default to heuristics: the person they’ve already heard of, the person a trusted colleague recommended, the brand name they recognize on the resume.

This is why The Talk matters. In an attention economy, every gimmicky tactic for getting attention eventually gets copied until it stops working. But relationships built on trust endure. A talk is one of the few things that gets better every time you do it, builds relationships with every audience, and positions you as someone with a point of view worth hearing.

You don’t have to become an influencer, or obsess over your “brand.” You just have to develop a talk.

If you need help, use my Custom GPT to come up with ideas. (The prompt is behind the paywall).

xoxo,

hils

Read more

My son turned 6 months old.

When I think back to the first few days with him, what I mostly remember is feeling like a protective film had been ripped off of me. One I didn’t realize I’d been wearing my whole life.

Have you ever used a device for years before realizing it had a protective layer you were supposed to rip off on day one? If you had asked, before my son was born, if I embraced vulnerability in my life, I would have said, sure. It doesn’t come naturally to me, but I try, and I’ve come a long way.

Having a baby made me realize the “long way” I had come was about a millimeter.

I was surprised to learn I was pregnant the day after Christmas. It felt like a miracle. We had tried for years to no avail. Failed IVF. It is a hell. But even hell can have its blessings. The struggle forced me to stare down the voices in my head that told me I was ambivalent about motherhood, declare them all a pretense, and abandon them. I no longer had the luxury of saying, oh, maybe it will happen, who knows. If I wanted a baby, we would have to actively work toward it. Which meant I had to unpack what I was so afraid of.

What was I afraid of?

Well, for one: everything. That a baby would never materialize, and the heartbreak would kill me. That something would happen to the baby, and the heartbreak would kill me. That the mental health I’ve spent my entire adult life cultivating would fall apart, and then what kind of mother would I be?

But surely everyone has these fears. Billions of women have faced them and come out the other side. Was I really more afraid than all of them?

Subscribe now

From the outside, motherhood always seemed to me like a turn inward. Women who were previously ambitious and engaged in the world suddenly become preoccupied with maintaining, at all costs, a nap schedule. They develop strong opinions about strollers. They vanish.

I was afraid of losing my ambition. Which is funny, because I never even liked my ambition all that much. The side of me that would be very happy weaving beaded bracelets all day has always found the ambitious side deeply inconvenient. But at least this was a duality I knew well, and the tenuous equilibrium between them had shepherded me into some remarkably interesting jobs. Dream jobs! Challenging jobs that fulfilled me creatively and intellectually. They were stressful, very stressful, but I had been bored in my work before, and being bored was worse.

My God, what if motherhood was....boring?!

Fear is weird. I was afraid that my ambition would disappear, and also afraid that it would prevent me from being the kind of mom I wanted to be. Would my ambitious side and my quieter side cancel each other out, or...finally merge? Might I start approaching the preparation of vegetable puree with the intense fervor I once reserved for Powerpoint?

It will not surprise you to learn that I was very in my head throughout my pregnancy. I may sound like an anxious person, but I’m not, not really. Pregnant me was a different story. I was anxious in ways I had never come close to experiencing before.

So imagine my surprise when my son arrived and ripped the protective film clean off. There he was. What now?

Victor Hugo’s original title for Notre-Dame de Paris was Ce qu’il y a dans une bouteille d’encre, or: what there is in a bottle of ink. I have always found this beautiful, and I’ve revisited it often in my head these past six months. When I look down at my son, I see a bottle of ink, full of possibility, of joys and heartbreaks to come. I understand now what I didn’t before. What might look to an observer like a turn inward is something far more profound. It is like seeing a blade of grass with a microscope for the first time. You realize that something you spent your whole life thinking was small contains the infinite.

“To see a world in a grain of sand...”

One thing I did not expect: I have never in my life felt more ambitious. Every minute is a bottle of ink, contains the infinite. The opportunity cost of my time is high. If I am to give up a single opportunity to hold my sleeping baby in my arms -- by far the greatest feeling I have ever experienced, and a joy I know is fleeting -- it absolutely better not be for some bullshit reason. And aren’t most of the reasons bullshit reasons?

I was prepared for a baby to make me softer, less available, less focused, less hungry. And maybe from the outside that’s what it looks like. But something happens when you find the thing you never want to trade: you stop tolerating the things you used to accept. You get ferocious, actually, in a way that would have embarrassed you before, because now there’s something real on the other side of every hour you give away.

This has led me to demand things of my time that would have previously seemed unreasonable, if not impossible. I’m in a position, for the first time, to actually try. I want to do work I love, that feels important, that pays me well, and offers me flexibility. I want to write and teach to help design a future of work where humans are not sidelined, because that’s the world I want for my son. I do not want this work to come at the expense of my ability to show up as the person I want to be for the people I love.

Fear has a way of narrowing what you think is possible. Like you have to choose between things that may not actually be tradeoffs, or like it’s safer to talk yourself out of your dreams than to lay the groundwork for them. The sentence you start today could become Notre Dame. I want my son to see the same expansive futures my parents showed me.

I spent my whole pregnancy bracing for my world to get smaller. Instead, the smallest thing in it blew it wide open.

My ambition shows up when I hold my son for one more minute before putting him to bed. It shows up when I sit down to write after he falls asleep. It was all in the bottle of ink.

xoxo,

hils

Subscribe now

A quick note before we get into it: The next cohort of my course, How to Become a Supermanager with AI, starts Tuesday. If you read this essay and think “ok, but how do I actually do all of this?” …well … that’s what the course is about! Building the tools, setting up the shared context, codifying your standards, driving adoption across your team. Enroll by midnight tonight and get 20% off.

There’s a conversation I’ve been having a lot recently which goes something like this:

“Exciting news, Hilary, I’m [flexes muscles] AI-native now. I taught AI agents how to do my job for me. I just tell them what to do, and weigh in at certain points to make sure the work is good. And hey, isn’t that basically management? I’m managing agents! And my agents are managing agents! Is this the future? Do we even need people managers anymore?”

First of all. This is certainly part of what management entails. But it is not the hard part of management. If “tell them what to do and check in at key points” were all management required, people would not cry at work as often as they do.

But I’ll take the bait.

Why do we need people managers in a world where AI does most of the work? Well, managing people and managing agents have a lot in common, and a good people manager will usually make an excellent agent manager. But the reverse is not always true, because management has never fundamentally been about doing the work. It’s about creating the conditions where good work happens.

To be clear, many managers are failing to do that today. It’s very hard to do well, even if you are trying your best. And not everyone is trying their best! I have certainly met managers who seem to exist primarily to gatekeep information and sit in meetings. These are not good managers and I think there will be less room for them to hide in the future.

But good managers do exist, it is possible to become one, AND we still need them even in AI-powered organizations.

Here are five core pieces of the manager job:

1. Own the outcomes

2. Create the shared brain

3. Build the team’s judgment

4. Decide what not to work on

5. Be the connective tissue

Let me walk through what I mean by each one, and how I see it changing with AI.

Subscribe now

1. Own the outcomes

“Getting work done” is about to get very easy. Agents can write the email, build the deck, draft the campaign, run the analysis. And they’ll get better at all of it. So the question becomes: if everyone can produce more, what actually differentiates a team that’s performing well from one that isn’t?

Outcomes. Did the work move the needle? Did it solve the right problem? Did it produce the result the business actually needed, or did it just produce a lot of stuff?

This has always been the manager’s job, but it used to be tangled up with execution. You were accountable for outcomes and you were involved in the work, so the two felt like the same thing. When agents take over more of the execution, what’s left is the part that was always the real job: making sure the work actually matters.

People sometimes fight about who should get to make a given decision. “Why does the product director get to make this call instead of the design director?” The answer is usually pretty simple: because it’s the wrong call, the product director is the one who gets fired. That’s what owning an outcome means. You’re on the hook for something you can influence but can’t fully control, and that requires asking the questions that sit above the execution. Are we solving the right problem? Is this the highest-leverage use of this team’s time? Are we optimizing for the metric that matters, or the one that’s easiest to move? And does the team even have a shared understanding of what we’re trying to achieve and why?

Orgs have a variety of things they need to do, and these sometimes conflict with one another. The growth team pushes for revenue; the product team pushes for user experience; the tension between them is what produces a good net outcome. Without somebody on each side who feels real ownership, who will push back and make trade-offs and lose sleep over the result, you don’t get that tension. You just get whatever got optimized first.

The more agents are doing the execution, the more important it is that someone is accountable for whether the execution added up to anything.

2. Create the shared brain

Good managers have always been in the business of context distribution. You’re in meetings your team isn’t in. You hear things about where the company is headed, about what leadership cares about, about what’s happening on other teams. And part of your job is figuring out how to get that information to your people efficiently, without overwhelming them or spooking them or turning every Slack message into a fire drill.

This is hard! I used to send my team a weekly-ish observations doc, just a running list of things I was hearing and thinking in my day-to-day that I wanted them to know. It was one of the most useful things I did as a manager, and it was also a pain to maintain.

Now imagine this problem in an AI-forward world. People on your team are teaching their agents how to do work. They’re giving them context, instructions, workflows. “Here’s how we write launch emails.” “Here’s what our brand voice sounds like.” “Here’s how we make decks.” I have what I call a “Manager OS,” which is basically a set of files and folders that contain everything my agents need to know to do work on my behalf: my standards, my preferences, how I want things done, key context about the team and the org. It’s great. But what happens when everyone on the team has their own version of this? Their own slightly different set of instructions, their own slightly different understanding of what “good” looks like?

It gets messy. And someone needs to be the editor. What’s the source of truth for the team? When someone figures out a better way to do something, how does that get into the shared pool so everyone benefits? (Who decides if it’s actually better?) When is information considered outdated, and what happens to that information? When there’s expertise on another team (legal, compliance, domain knowledge) that should be baked into how your agents work, how does it get brought in?

This was always the manager’s job. But it used to be about verbal updates and shared docs. Now it’s about maintaining the actual context directory your team’s agents run on. The stakes are higher, the surface area is bigger, and the consequences of getting it wrong are that your team is producing work based on bad information, at scale, all day long.

(Mechanically, maintaining a shared brain/context directory/folder full of markdown files is currently a pain. I know folks who do it in Github or OneDrive or Google Drive. A ton of small and big companies are working in this product space so I expect it to get much easier in the next 6 months. But the art of curating it well will remain a competitive advantage).

3. Build the team’s judgment

Here’s a question that has always been central to management: do I have the right people, and are they getting better?

Pre-AI, this meant assessing your team’s skills, identifying the gaps, and figuring out how to close them. Coaching. Training. Pairing people with mentors. Giving them stretch assignments that would force them to grow.

This is still the job. But I think we’re underestimating how much harder it’s about to get. When agents do more of the execution, people get fewer natural reps. A marketer who’s using AI to draft campaigns can move faster, but do they understand why one version works and another doesn’t?

That judgment comes from doing the work. And if agents are doing more of the work, the question becomes: how do you make sure your team is still developing that sense?

That’s the manager’s job. You’re the one responsible for making sure people are actually getting better, not just producing more. That might mean slowing down sometimes. Saying, “Don’t use the agent for this one. I want you to do it yourself so you understand how it works.” It might mean building tools that teach rather than tools that just do: a tool that reviews someone’s work against your criteria, or one that walks them through your framework for making a decision and explains its rationale as it goes. Every time they use it, they’re getting a rep. They’re learning what good looks like, in context, at the moment they need it.

But somebody has to care about this. Somebody has to know what the team’s gaps are, decide what good looks like, and create the conditions for people to close those gaps. The team won’t optimize for their own development; they’re busy shipping. The manager is the one with the wider view and the longer time horizon.

4. Decide what not to work on

When everyone can move faster and produce more, a new risk emerges: too much gets done, but none of it matters. As I say often, it’s never been easier to build, which means it’s never been easier to run 10x as fast in the wrong direction.

The manager’s job is to be opinionated about where attention goes. What are we working on and why? What are we not working on? Where does the team need to slow down and think?

This gets into a genuinely tricky tension. Should someone spend time learning AI tools if it means their work doesn’t move forward today? Probably yes, actually, but that’s a hard call for an IC to make on their own, because they’re measured on the work. The manager is the one who can say, “I’m giving you a day to automate part of your workflow. I’ll provide the air cover. Your work can wait.” That’s a prioritization decision that requires someone with a wider view.

And as the number of possible things to work on explodes (because AI makes everything possible), the question of what to actually do becomes the most important question on the team. The manager is the one who has to answer it.

5. Be the connective tissue

There’s a reason the cliché about managers is that they’re “always in meetings.” It’s because a huge part of the job is being the connective tissue between your team and everyone else.

Across teams: making sure your team’s work fits the bigger picture, catching the conflicts before they ship. “Hey, did you know marketing just launched something that contradicts what we’re building?” Across levels: negotiating for resources, making the case for your team’s priorities, translating what leadership wants into something your team can actually act on. And up: making sure the people above you understand what’s happening, what’s working, what’s stuck, what they should be worried about.

You might think some of this gets easier with better tooling. Everyone, and I mean everyone, wants to automate “updates.” But organizations are not static environments. Teams are constantly changing their approach, learning, responding to competitive threats. The landscape shifts every week and the coordination problem shifts with it. Somebody has to be reading the room across those boundaries, building the relationships, and making judgment calls about what your team needs from other teams and what they need from yours.

There’s a reason managers have to work well with others. It’s because the job is others.

So what’s the point?

I think the reason this question keeps coming up is that a lot of people have only ever experienced management as execution oversight. And if that’s all it was, then yes, AI would replace it.

But the managers I’ve learned from and tried to be were never primarily doing that. They were owning the outcomes, create the shared brain, building judgment across the team, deciding what not to work on, and stitching the whole thing together across teams. That work has always been hard to see and hard to measure. AI is just making it impossible to ignore.

And it makes it more important than ever, because the cost of getting it wrong now scales with the speed of execution. Bad context, applied at scale, all day long. A team that produces more but learns less. Speed pointed in the wrong direction. These are expensive problems, and they are all management problems.

xoxo, hils

More Writerbuilder

The next cohort of How to be a Supermanager with AI starts in one week — now in a Europe-friendly time slot! Here’s 20% off through Friday.

Over the weekend I saw a headline that Anthropic “launched a new tool to migrate ChatGPT memory to Claude.” That’s interesting, I thought. How would that even work? ChatGPT doesn’t have an export-your-memory feature. There’s no API for it. Is Anthropic somehow reaching into OpenAI’s servers and pulling your data out? That would be... illegal, probably?

Well, it turns out the “tool” is simply asking ChatGPT to tell you what it remembers about you and then pasting that into Claude.

That’s it. Copy and paste.

This was a very clever, well-timed, and well-executed move by Anthropic. But is this a tool? Is there a migration or an export happening here? I don’t think so.

That’s not an insult. Some of my best friends are Copy and Paste! When actual technical people watch me work with AI they are always flabbergasted by the amount of copying and pasting that I do. “There’s an MCP for that,” they say. “I don’t care,” I say. Frankly, I am too lazy to set that up, and I am scared of letting my tools hang out without me. What will they get up to? Will they talk about me? I don’t intend to find out.

I bring this up because I just wrapped the latest cohort of my Supermanagers course, and the #1 question I got from students was some version of: “I get that you can do cool stuff with AI, but my company won’t let me connect my tools to the data I need, so I’m struggling to make progress.”

This is a real frustration, and I get it. It feels like you’re being held back by bureaucracy when all you want to do is build something useful.

But my answer, which I ask with love: did you try copying the data and then pasting it? Did you try taking a screenshot?

I know, I know. I used to feel bad about asking this, like I was minimizing the challenge. But now I am emboldened by the fine team at Anthropic who have legitimized this approach on a national stage.

It’s tempting to think that “access to data” will magically solve you problem. If you could only connect your CRM or analytics dashboard or project management tool directly to your AI tool of choice, all your problems would be solved. You’d finally be able to automate all your busywork.

But in my experience, if you can’t get value from AI with copy-paste, you are going to struggle to get value from a fully integrated solution.

Let me play this out. You want to plug your Google Analytics into an AI tool so it can send you a weekly insights report. You get access, it runs, and the first report flags that signups dropped 20% last Tuesday. Except you already knew that. It was a pricing page test you ran. So now you need to figure out how to teach your tool the difference between signal and noise.

That’s a management problem as much as it is a technical one. And it’s one you would have surfaced and figured out how to fix if you had started by pasting screenshots of your data into the tool.

Instead, now you have a broken tool that’s causing fire drills for people who don’t have the context to know the alert is meaningless. You know who wouldn’t have let that happen if they were doing a good job? The person who is responsible for data integration permissions at your company.

That’s right — the people who are making it difficult for you to get the data you need are doing their job well. They’ve been around the block. They know there is a delta between someone saying “I have a really good idea but I haven’t really thought it through” and that idea actually being a net benefit in execution. They should interrogate your plan. Unfettered access to data creates problems that other teams have to clean up.

But hold on. I’m not going to take their side either. Because forcing people to write up a perfect business case is a waste of time. Some trial and error is necessary to build tools that are actually good. You can’t perfectly scope an integration you’ve never experimented with, especially if you are new to all this.

So what’s the solution?

SCREENSHOTS, BABY. MANUAL ONE-TIME EXPORTS. COPY! AND! PASTE!

Build a janky version of your idea as a proof of concept. Figure out 80% of the issues. Use it for a few days to validate that it is, in fact, as useful outside your head as it is inside your head.

If you want something, prototype it. And if you have a working prototype — even a terrible one held together with screenshots and good intentions — it becomes a thousand times easier to secure the permissions you need to make the thing real. You might even secure resources. A team!

But you are not going to secure those permissions or those resources if your pitch is: “Can I please have read/write access to our database that I don’t fully understand for reasons I haven’t totally figured out?”

A better pitch: “I built an agent that proactively notifies the team when there’s an unexplained drop in usage in one of our features. It actually flagged that Feature A usage had been slowly eroding since January, recognized it might be related to an upstream UX change we made, and proposed a fix. Right now this thing is held together by a string of manual screenshots. I don’t have time to do that anymore so it’s going to stop working. Can you help me get read access to the usage analytics API so it can run automatically every morning?”

Subscribe now

See the difference? The second pitch works because it does three things: it proves the tool solves a real problem (not a hypothetical one), it already works (you’re not asking someone to imagine the value — they can see it), and it has a clear failure mode if you don’t get help (it’s going to stop working, which means the team loses something they’ve started relying on).

Don’t get caught in a chicken-and-egg battle of “I need the data to prove the value, but I can’t get the data until I’ve proven value.” There’s always a way to prototype around it.

“But Hilary,” you say, “it’s not that simple. My data is…sensitive. My industry is…complicated. I can’t just go putting it into an AI for fun and profit.”

OK. Fair. Here are your options:

Blind it. Strip out the identifying information and work with the structure.

Find a proxy. There may be a similar, publicly accessible dataset that’s close enough to build with.

Or, my favorite, ask the AI to generate fake data that is formatted like the data you wish you could play around with. Describe what your data looks like. The fields, the patterns, the messiness. Then go nuts. If you are able to design a tool using that fake data that will actually transform the way your team works, you will have absolutely no trouble getting someone to sit down with you and figure out how to make it work with the real thing.

I’m always telling people they should simply make fake datasets. I think this is vexing to them. “What good can a fake dataset do? I can’t make decisions with a fake dataset.” Sure. But presumably you don’t have the dataset you need. Prototyping what you’d do with it is a great way to get that access.

Anthropic, a company with some of the best engineers on the planet, looked at the problem of moving your data from one AI to another, and their answer was: just copy and paste it. They shipped the simplest possible version, and it worked. You can do the same thing. Start with what you have. The key is to just start.

And if you need synthetic data to get started, I’m including some prompts for generating it below the paywall.

This is the kind of stuff we dig into in my AI for Managers course: how to actually put AI to work for you, how to manage AI agents the way you’d manage people, and how to stop waiting for perfect conditions and start building. The last cohort gave it a perfect 10/10 rating across the board (!). A few things students said:

“I thought I knew how to use AI well + get impact, this course made me realize how little I actually knew and opened my eyes to approaching AI in a completely different way.”

“My team has benefitted from this course, not just me!”

“You not only learn the concepts, but actually apply them and walk away from the course with real usable tools that you have built.”

The next cohort starts Tuesday. I doubled the length and content this time around to ensure it stays up-to-date with the cutting edge. Here’s 20% off enrollment through Friday.

Read more

Quick plug: Anjali Ahuja and I are doing a free Lightning Lesson on how AI is changing the PM-engineer dynamic, in partnership with Lenny’s Newsletter. Anjali invented the AI PM role at WHOOP and has a ton of great insight to share. Grab your spot here!

Every leader I talk to asks me the same question: how do I get my team to actually adopt AI in a non-superficial way?

It turns out that this is quite hard. And it’s hard because you need a bunch of people who have both technical expertise — the ability to identify load-bearing business problems, scope the right solution, and build it — AND company-specific domain expertise: how the work gets done, what makes the work good or bad, what obstacles prevent good work.

Subscribe now

This is, in part, why engineering orgs have been much better at transforming how they work than other functions. They are not merely technical; they understand their own obstacles and they understand how to deploy software against them.

If you want other teams to radically change how they are working with AI, you need to find a way to either get the domain experts sufficient technical expertise, or get the technical folks sufficient domain expertise.

My favorite example of the latter comes from Notion, where they embedded an AI software engineer on the sales team for a month. He carried a quota! He made 40 cold calls on his second day and booked 0 meetings! This is art. I am obsessed. The next time an engineer describes themselves as “curious” to me I am going to ask them if their curiosity is strong enough to endure 40 cold call rejections in one day.

He ended up creating a new outbounding tool that the reps describe as “useful” (!). And on the other side of the spectrum — a sufficiently technical domain expert — a designer at Notion built a collaborative “prototype playground” where the entire team can create, share, and iterate on functional prototypes. He saw that designers wanted to do more code-based prototyping but were afraid of breaking things. And fixed that problem! With software!

I’m sure every CEO would love for their employees to embody the extreme skill malleability on display at Notion. But what if they don’t? How do you solve the lack of technical / domain expertise overlap?

One possibility: send in the consultants. Bring in the outside engineers, shake things up, hand over the playbook, move on. And I’m sure this will work, somewhat. But there’s a problem with this approach.

If you want your team to embrace AI, like really embrace it, you probably don’t want to give them the impression that you are foaming at the mouth in anticipation for the day you can finally fire them all and retreat to your Agent War Room to lead your army of robots in peace.

You know what signals to your team that you think they suck at their jobs and are also a pain to deal with? Bringing in a bunch of sharks to critique their work, sniff around for inefficiencies, and extol the virtues of their Vertical Playbook™.

You know who is not going to be super keen to rally behind their leader and “get with the AI program?” … You see where I’m going with this.

Even setting aside the morale problem, the consultant probably can’t do this well either. I’m not wholly anti-consultant (I’m married to one!!!). There are plenty of times where a macro lens + outside perspective is needed. I just don’t think this specific challenge qualifies. The whole problem is that you need deep domain expertise alongside technical skill. So if you’re bringing in a team to overhaul your marketing org, you would want them to be as good at marketing as the best marketing agency AND as technically sharp as, I don’t know, Palantir? And also understand your company’s idiosyncrasies, strategy, etc? And win the trust of your team? It’s the same problem we started with: hard-to-find skill combos. Maybe a few such firms exist and I’m sure business is booming for them. But it feels a little too good to be true.

I understand the urgency, and the feeling that outsiders may be better equipped to cut through the status quo slog than the people desperately clinging to their calcified jobs. And I admire leaders who can do the unpopular but necessary things to radically change the company’s course. Extreme paranoia in a CEO can be a superpower and you have to take the bad with the good.

But. BUT!! I also think this approach fundamentally misunderstands what it will take to succeed in the AI era. My biggest takeaway from teaching hundreds of people AI skills is that learning to fully leverage AI requires a rewiring of your brain. You start thinking about your work, your skills, your time in a totally different way. I think this is particularly true for nontechnical people, who are currently speedrunning the lessons that senior engineers spent their whole career learning.

In my course on AI for managers, we build a lot of tools. Students are always eager to build something that is “actually useful” right off the bat. But this is hard to do! Many decorated product managers have spent their entire careers failing to build something actually useful. It requires a honing of several different types of intuition: how to identify the right problems, how to scope them, how to pick the right solution from dozens of possibilities, and how to iterate toward something that works. So I tell them not to worry about this right away. The early building is more about training your brain to think in a completely new way. Eventually you will reach a point where your team will tell you they don’t trust the marketing attribution methodology and you will think, “I can build something that will solve this.” But you must cross the desert of bad apps first.

So this is why I think the best way to actually drive ~AI transformation~ at your company is to invest in helping your domain experts (that is, your team) develop sufficient technical skill to redesign how they work. They develop this skill by doing the work themselves, not by having some outsider do it for them and then handing them the keys at the end.

They need their brains to change, not just their tools.

How do you help domain experts develop technical skill? A few hours of training will not do it. You need to:

Give people the time to experiment. This is the big one! Most people are simply too bogged down in their jobs. Block off two days and challenge them to build a tool that will make their work better. Do it again next month.

Make it clear that transforming how you work is actually a priority. Don’t tell people that something else is the priority and then get mad when they spend their time on that other thing instead of this. Is ~AI transformation~ one of the top 3 company goals this year? If not, it’s not actually a priority. If it’s not a priority, it’s not going to work.

Set cartoonishly ambitious goals. Don’t tell your team that the goal is to find a reasonable middle ground on AI. If you are hedging, your team will hedge ten times as much. Speak in absolutes, like, “I never want to read another document again; I will only accept prototypes.” Make them feel the urgency.

Visibly reward wins. Did someone on your team build something cool? Make them feel like a million bucks. Tell everyone about how cool it is. Ask her to demo it to the team and share how she built it. Invite her to the secret cool meeting where all the important decisions happen. People notice what gets rewarded, tolerated, and punished, and this shapes their behavior.

Build a culture of learning. Let the team experiment. Some things won’t work, especially at the beginning. Ask them what they learned from the experiment. Challenge their thinking. Ask them what they’ll do differently next time. Hold them accountable to learning and getting better. “Trying stuff” is not the same as learning.

Become an expert yourself. There are so, so many resources available to people trying to learn these skills. Find the best resources and share them with your team. Model what you want to see! If you can’t carve out the time to do this, don’t expect them to.

Maybe you are concerned your people don’t, as I like to say, have the sauce. I will admit that not all teams are capable of transforming themselves. But it has nothing to do with where they are starting. All that matters is that they are good learners. If they can learn new things quickly, and if they have any aptitude for self-directed learning at all, they will figure this out. If they don’t… are you sure? If you’re sure, fine, hire consultants.

But if you give your team the time + resources + clarity + accountability, not only will you find yourself with employees that Notion may try to poach from you, but more importantly, you will demonstrate to your employees that you believe in them and are investing in their future careers. Bringing in the consultants signals “I am about to fire you.” You should not be surprised by a lack of enthusiasm from people who believe they are training their robot replacements.

And look, I’m not stupid. While I genuinely believe that long-term human-robot collaboration is both possible and economically preferable to robots taking over everything, I understand that even in that world there will probably be fewer human jobs. You can’t pretend your ~AI transformation~ is not at least partially motivated by a desire to trim down the organization. If some people have to go, OK, but you can at least try to upskill them on the way out.

Do this right, and you might be surprised. Once they get over the learning curve and the rewiring kicks in, you may find models for keeping humans in the loop that are better than the fully automated supercompanies we are being told are inevitable.

I’d like to see a future where humans aren’t fully sidelined. If you want that too, don’t start by sidelining your own people.

xoxo,

hils

Today, I’m introducing a paid tier to my newsletter. Everything I write (essays, ideas, analysis) will stay free. The paid tier is for the specific prompts and step-by-step instructions behind what I’m writing about.

Subscribe now

I also opened a new cohort of my Supermanagers course, starting March 10. I rebuilt it from the ground up, and it now covers everything from prompting fundamentals to building an agentic management system. Enroll here!

---

Way back in 2012, I attended a talk that has rattled around in my brain ever since. At SXSW, the New Yorker cartoonist Matthew Diffee gave a talk on creativity titled “How to be an idea factory.”

For a single-panel cartoonist, the gap between idea and execution is quite short. You have an idea, you draw it, done. Because of this, cartoonists can afford to pursue hundreds of ideas in a week. Maybe one or two of those will get published. This makes cartooning into something of a numbers game: to have good ideas, you simply need to have as many ideas as possible. Which means to be a successful cartoonist, you need to figure out how to turn your brain into an idea factory.

His advice for aspiring creative people: “Be like a mother sea turtle.” Lay a hundred conceptual eggs in the sand, swim off, and never worry about them again. Your job is not to fret over the eggs that don’t hatch, or the hatchlings that get eaten by predators. Just to keep laying eggs.

Now, as a New Mom™, the idea of this hapless sea turtle makes me much sadder than it did nearly 15 years ago. But, I agree that the secret to having good ideas is having a lot of ideas. The problem is that if your work requires time and effort to execute, you can’t simply fling ideas off into the sea and move on. You need to commit! And out of a hundred ideas, maybe two of them are really worth pursuing.

….So how do you know which two those are?

Stop doing more

This recent HBR article finds that AI tools don’t reduce work, they consistently intensify it. This lines up with what I’m seeing. The initial thrill of “look how much I am capable of doing now!” quickly gives way to “oh no, I can’t (or don’t want to) keep up with this new pace of work.”

I immediately stretched myself too thin after leaving my job. Teaching, writing, speaking, consulting...it was tempting to say yes to it all. So I did! And AI helped me stay mostly above water. But time is still finite. Focused attention is still finite. I could feel myself becoming overcommitted and overwhelmed. My lack of focus was starting to bear a real cost.

AI makes it very easy to do more, more, more. But why do we want to do more? Most work fails to have the impact we hope it will, which means that most people are making mistakes most of the time. Why would we want to make even more mistakes?

Well, in theory, because we can’t predict how things will play out. We don’t know which ideas will hit! And then we need some way to manage all this work, which of course creates more work, and oh, maybe AI can help with that too.

But what if instead of enabling us to do more, we used AI to help us make smarter decisions about what to work on?

I love Peter Yang’s response to everyone sharing their AI-powered task management tricks: “Why do I need AI for this? I only do 3 things a day. I write them on a whiteboard. Then I do them.” Peter has a full-time Product job while also running a successful newsletter, podcast, and community. That’s the power of focus.

Before AI, maybe you had 10 ideas about what to work on, you picked 5 to experiment with, and ended up focusing on 3.

With AI, you might have 100 ideas about what to work on, so you pick 50 to experiment with, and focus on 30. Because why not, right? The cost of building is zero!!

Except, of course it’s not zero. If you’re splitting your time across 30 projects, each one gets ~3% of your attention. You’re diluting your best ideas.

Which means the question to ask in our AI times is not “how do I do more?” it’s:

1. How can I have 10x more ideas than I had before? (increases the likelihood of having a good idea)

2. How can I test them rapidly so I can make better decisions about where to focus?

Ultimately, you should still probably be focusing on 3 things max each day. The “AI unlock” is in making sure you’re picking the right things to focus on.

I talked a bit about my process for doing this in a recent Lightning Lesson with Tal Raviv and Aman Khan.

Make space for ideas

First of all, it’s very hard to come up with good ideas if your brain is churning away nonstop on urgent-feeling bullshit. It’s important to always be a little underutilized.

Alex Soojung-Kim Pang’s book Rest: Why You Get More Done When You Work Less makes the scientific case: your brain does its best creative work during periods of rest, not labor - consolidating information, making novel connections, linking disparate ideas.

In his SXSW talk, Matthew Diffee referred to a similar concept as “flipping the funnel.” When it’s time to stop gathering fodder and start creating, take the imaginary funnel over your head, the one mainlining information from TV, social media, podcasts, newsletters, and invert it. As in, wear it like a helmet. Protect your head from all that nonsense. Give your thoughts some space to get weird.

I aggressively protect time and space for my mind to wander. I rarely listen to music/podcasts while walking/driving/puttering around (yes, I drive in silence, I know this is psychopath behavior, sorry, I have a rich inner life!). The point is you can’t “productivity hack” your way to having more ideas. You have to give them space.

Never let an idea slip away

I am constantly having ideas and then immediately forgetting them. This wasn’t a big deal when I lacked the time to pursue most of my ideas. But now, every idea is something I can quickly pressure-test and possibly pursue. The leaky sieve of my brain is suddenly expensive!

So I set up a simple system to capture every idea I have, whenever I have it (which is usually when I’m falling asleep and I don’t want to open my eyes).

I have a Wispr Flow shortcut on my iPhone Lock Screen. Any time I have an idea, I tap and talk. On walks, in bed, in the middle of the night when I can’t sleep and something hits me. Importantly, I don’t have to open my eyes or be looking at my phone to use it.

It’s very easy to set this up on an iPhone:

1. Create the shortcut: Open the Shortcuts app > New Shortcut > search for Wispr Flow > select “New Wispr Flow Note.” Make sure it’s the first shortcut in your list. (You need to have the Wispr Flow app installed for this to work).

2. Add it to your Lock Screen: Go to Settings > Wallpaper > Customize Lock Screen > Add Widgets > select Shortcuts. Tapping it from your Lock Screen will trigger the first shortcut in your phone’s list of shortcuts.

You can also use AI to dig up ideas you’ve already had that have been lost to the sands of time. Ask Claude Code to dig through your screenshots, your Twitter likes, your notes app, or wherever you might have “saved” and then immediately forgotten an idea.

Test rapidly, then commit

I’ve set up a system that helps me keep my ideas organized, and helps me test them out.

My claude.md file has a note (behind the paywall) that tells it to look out for a variety of different ideas in our conversations -- newsletter ideas, product ideas, lesson plans -- and save them in the appropriate file. Then, I have a variety of prompts I tag in that refine my ideas into something more substantive that I can test.

For example, I recently had an idea for a “to-don’t list,” basically the inverse of a to do list. Here’s the extremely rough, stream-of-consciousness idea I captured via my Wispr Flow shortcut:

So there’s that book “4000 Weeks” where the premise is basically productivity is a scam because the faster you do tasks, the more tasks have come to you. Every time you knock out one task, two tasks appear in its place. So the key is to actually accept that time is finite and learn to do fewer things, or to say no to more things. I want to better apply that in my life. So I want a to-do list manager, but I want it to be a to-don’t list. Anytime I put something on my to-do list, I want it to talk me out of it and try to convince me that it’s not actually important enough to do, and then help me get out of whatever obligation I have to do it.

I use my “idea refiner” prompt (behind the paywall), which asks me a few questions and then generates a spec for the idea. Then I use my “validation approach generator” prompt (behind the paywall) to come up with five different possible implementations. It gave me:

1. A “pushback partner”: you tell it what you’re about to say yes to, and it tries to talk you out of it

2. A “no coach”: a quick-access prompt on your phone for when you’re in the moment and tempted to commit

3. A “regret inventory” that looks backward at what you already did to assess which commitments you regret vs. are glad you did

4. A “one thing to drop” morning ritual, it picks one thing on your plate and makes the case for dropping it

5. “decline drafter”: you tell it what you need to get out of, and it writes the email

The AI writes a prompt for each of these ideas so I can test it out myself for a week and see which one, if any, are actually useful enough to share with other people. (Watch the Lightning Lesson for a demo of this).

I tested all five, turned my favorite into a Custom GPT, and shared it with my newsletter list. It got used about 100 times. My most popular GPT has been used tens of thousands of times. So that’s a kill -- it didn’t get the traction I thought it might.

This is how I arrive at a lot of the Custom GPTs I make. I think of them as basically prototypes for some bigger idea I have. They are very quick to make, so I can send them out through my newsletter and see which ones actually get traction. If one is particularly popular, I start thinking about how it might turn into a more durable product, like a workshop or a course or even (gasp!) a piece of software. And then I focus on executing that idea well.

The foundations haven’t changed

To have a good idea, you need a lot of ideas, and you can’t know in advance which ones will work. That hasn’t changed. But now, you can test an order of magnitude more ideas in a very short amount of time.

When people spin up a bunch of work for themselves, they are hedging. If you don’t know which idea is the idea, you try a little bit of everything. This is not generally a recipe for success.

Work has diminishing returns. I don’t want the AI to help me do more tasks that accomplish nothing, and I certainly don’t want it to trick me into believing that I have somehow outsmarted statistics. I want it to use its infinite intelligence to help me make better decisions about where to direct my focus. That’s a model of human-robot collaboration I can get behind.

xoxo,

hils

Behind the paywall:

1. The Claude.md instructions I use to capture and organize my ideas

2. The product idea refiner prompt

3. The validation approach generator prompt

Read more

1. Supermanagers closes tonight

How to Become a Supermanager with AI starts tomorrow, and enrollment closes tonight at 11:59 PM ET.

If you’ve been on the fence: this is the course where you learn to build AI tools that scale your expertise! Give feedback the way you give feedback, coach the way you coach, help your team do better work without waiting for your calendar to open up.

I will also be revealing my souped-up “Manager OS” for the first time, along with how to build your own. But don’t let that scare you. The course is super accessible regardless of your skill level, and most importantly it’s lots of fun :)

150+ managers from companies like Google, Microsoft, Visa, Dropbox, JP Morgan, Khan Academy, and NBCUniversal have joined, with a 4.7/5 rating.

Enroll here before midnight. Maven has a guide here for getting the course reimbursed by your employer.

Reach out with any questions!

2. Free Lightning Lesson: How to Know What AI Products to Build

Next Monday (Feb 9) at 1 PM ET, I’m doing a free session with Tal Raviv and Aman Khan, authors of the very very excellent guide to using Cursor to build AI Product sense in yesterday’s Lenny’s Newsletter.

The session is about how to come up with and validate AI product ideas: turning friction into testable products, distinguishing novelty from lasting value, and filtering for retention. Whether you’re building side projects or managing a roadmap, this one’s practical.

670+ people already signed up. Grab a spot here.

3. I was on The Economist’s Boss Class podcast

The Economist launched a the third season of their podcast called Boss Class about management, and it’s all about AI. I was a guest on the latest episode: The Easy Button.

We talked about using AI to raise the bar at your company: what’s working, what’s not, and what separates the managers who are actually getting results from everyone else still experimenting. Also Anna Karenina.

Worth a listen if you’ve got a commute this week.

4. I’m speaking at the Charter AI Summit (free to watch)

Next Tuesday (Feb 10), I’m speaking at Charter’s Leading with AI summit in NYC. My session is at 1 PM ET with Brandon Gell (COO of Every) on how AI power users are going beyond the basics.

The full lineup is pretty great: IBM’s CHRO, OpenAI’s Chief Economist, the CEO of Klarna, leaders from Microsoft, McKinsey, Dropbox, and more.

Virtual attendance is free. Register here.

Big week. Thanks for being here.

xoxo,

hils

[Quick housekeeping note: enrollment for my upcoming AI for people managers course closes in two days. Reserve your seat to learn from me, someone in the “top 1% of managers worldwide” according to The Economist, yes that’s a real quote.]

I was sitting in the window seat, homebound for Thanksgiving from SFO to BOS, thinking about how dramaturgically confusing it is that Mamma Mia 2: Here We Go Again uses some but not all of the same songs as the original Mamma Mia, when the kid sitting in the middle seat turned to me with panic in his eyes. “Can you help me?” he asked. I blinked at him, confused. “I took an edible,” he sputtered, “and I am too high and I am FREAKING out.”

Subscribe now

Without thinking, I sprung into action. Looking in his eyes, I said, “Hey, you’re safe. You are not in any danger. I just need you to breathe with me, okay?”

And so went the rest of the flight. (At one point, I suggested he try listening to some calming music, but he told me his music was stressing him out. I asked what he was listening to, assuming it was, I don’t know, club bangers. He told me it was James Taylor. I was like, “oh boy, if James Taylor is stressing you out, music is not for you right now.”)

When the flight ended, I walked with him to baggage claim. I saw a woman eagerly waving to him. “MOMMY!” he called out. I slunk into the shadows of the baggage carousel and disappeared.

This experience made an impression on me, though probably not for the reason you expect. At the time, I was deeply ambivalent about having children. I primarily worried that I did not have the “motherly intuition” you hear so much about. I would freeze up when friends handed me their babies. How do you hold these wiggly little guys? What if I accidentally hurt them?

But here, look! Irrefutable evidence that I did possess some level of maternal instinct. This kid needed me and I stepped up to the plate. I didn’t even need to think about what I was doing, it just came naturally to me. Surely that’s some Grade A Motherly Intuition.

Except, it wasn’t. It had nothing to do with mothering. Minus the drugs, I’d been in this exact same situation before. One of my best friends from high school suffered from terrible panic attacks. I was with her when she had one mid-flight, coming back from vacation. Fortunately, she has always been a strong mental health advocate (more on her down below, she rocks). She taught me everything I needed to know to support someone through a panic attack: how to focus them on their breath, how to reassure them that they are not going to die.

The reason I did not have to think about how to respond was not because I had some innate truth deep inside of me. It was a skill I had learned and practiced!

I think about intuition a lot: what it means, how we talk past each other on it, and why it can counterintuitively undermine our confidence in our own judgement.

You hear it used it two different, almost opposite, ways.

First, there’s the more feminine-coded version of intuition. A mother’s intuition, trust your gut, “follow your heart mama!!” and all that. This idea that if you tune out the distractions of the world and connect with the light within, you will simply know what you need to know.

And then there’s the more...maybe not more masculine-coded, but at least more professional-coded version. Think about “product intuition,” essentially the ability to predict how users will behave, or someone who has good marketing, technical, or hiring instincts. They may start out with some natural aptitude, but it’s understood as a skill that gets stronger with reps. In this framing, you develop good intuition by taking action: run experiments and update your judgment based on what you learn.

The problem is that we conflate these two things. People who have honed their judgment over time tell others to “trust your intuition,” because their intuition is reliable. But it’s reliable because they built it! And so people on the receiving end of this advice (often women, tbh) end up believing they’re lacking some important part of themselves. Something that should be guiding them, but isn’t.

This is a real bind. You feel lost, but you also have no framework for figuring out when you should trust your gut and when it might lead you astray. So instead of trusting yourself to figure things out, you just end up trusting your internal critics, who can be quite loud. And mean!

I felt this myself, feeling like my lack of natural “baby skills” was some kind of sign that perhaps motherhood wasn’t for me, rather than a perfectly reasonable result of being the youngest sibling and cousin in my family, who simply never spent enough time around babies to figure out their wiggliness. And I also see it everywhere, particularly on the various mom Subreddits, which are awash with women who have convinced themselves that their inability to correctly guess what their baby needs implies some fundamental failing that reveals they never should have had kids in the first place.

While I was writing this post, I happened to read Allison Bornstein’s latest post about starting her denim collection over from scratch postpartum. (I genuinely cannot imagine a topic more squarely pointed at my current interests). She talks about trying her old jeans on, and while they technically started fitting her again, certain pairs did not work for her new body. She refers to this process as “intuitive”: “I mean when you put something on your body, you immediately know whether it works or not. I think when we need to hem or haw, we are actually often trying to talk ourselves in or out of what our gut is telling us. Even though I felt a bit unstable in my body, I still have that intuition of what feels right to me.”

I don’t want this to sound like a criticism, because, there is literally no reason she needs to dissect the precise meaning of words (that is my job, apparently), and also her writing is excellent and I recommend it! But even in this very piece, she acknowledges that she worked as a “Fashion-with-a-capital-F” stylist for 10 years before moving into normal people fashion styling six years ago. I have no doubt that she started out with an above-average aptitude for fashion, but that’s a lot of reps! That is an extremely honed and practiced intuition!

So, yes. It’s intuitive in that she does not need to think about it. She “just knows.” But her gut is going to be a lot better at this than my gut. (I know for a fact, because my gut is also trying on jeans in a new postpartum body, and I have searched my soul, and there are no answers to be found.)

Denim issues aside, I was surprised by how much I enjoyed early postpartum days with my son, given my lack of skills. But I shouldn’t have been. I love being really bad at something and then rapidly getting better. I am at my best when scaling a steep learning curve. (I also like writing little songs where I change the lyrics of existing songs to be about having a clean butt, which was at least 30% of it.)

I was genuinely amused by how wrong my instincts could be. Like, how do you get a newborn to go to sleep? I don’t know... tuck him in with a bunch of stuffed animals? Nope, that could not be more wrong / illegal! But we tried things, we learned, we gradually got better, and our confidence improved. That was super empowering.

So, yes. “Follow your heart, mama!” and all that. But beware of how this advice can be a little misguided. It tells you to trust something you haven’t built yet, which can make you feel broken when it doesn’t work. The most useful thing someone could have told me before I had a baby wasn’t “trust your instincts.” It was “you don’t have instincts yet, but you’ll build them.” That would have freed me from years of wondering what was wrong with me. “Trust yourself” is good advice, but it’s the end of a process, not the beginning. It only makes sense once you’ve done the work of being wrong and noticing.

xoxo,

hils

p.s. what if instead of “follow your gut,” we said, “follows your guts”? I don’t think that would have caught on, lol.

p.p.s my friend Sami who taught me how to help someone through a panic attack is now a star clinical psychologist. You can check out her site here. When I sent her this piece, she told me, “I love this piece because it reminds me of what I try to help my clients build. So many of them don’t trust themselves and think that they automatically should even though they’ve been subjected to traumatic environments or chronic and validation are both. So a lot of my job is just helping them, build the skill to actually learn how to trust their gut or their inner wisdom which is exactly what you and many other women have to do as moms. This is voice to text so there may be some typos.”

Subscribe now

Read more Writerbuilder:

If you’re job searching right now, the system probably feels broken. You’re tailoring your resume, writing thoughtful cover letters, doing everything “right”, and hearing nothing back.

Last week I did a Lightning Lesson on how to actually get noticed in 2026. I’ve been on the hiring side for years and have interviewed hundreds of candidates, so I’ve seen what works, what doesn’t, AND how much it’s changed recently.

You can watch it here or see the key takeaways down below.

Subscribe now

But I’ve also long been frustrated by the feeling that simply telling people what I know is not enough. It’s hard to internalize lessons that other people have learned, let alone implement them. This is true for job seekers watching a career talk. It’s also true for your team sitting through a training!

I’ve been experimenting with turning writing into tools (see: writing is building now), and couldn’t resist the opportunity to turn our talk into something more interactive.

So, I used the transcript to build a series of job search tools. You can find them here.

Here’s the gist of the advice we shared in the talk, plus a few additional thoughts:

1. Proof of work > polished resume. In 2026, everyone’s resume looks good. (Thanks, AI.) What stands out is making something that shows how you think. Depending on the job, that could be a teardown, an analysis, or a prototype.

2. Know your “least charitable read.” Hiring managers pattern-match in a few seconds. If your background doesn’t obviously fit the exact background they are looking for, assume they will fail to connect the dots. Your job is to figure out what those assumptions are and get ahead of them. (I think this is especially important for career-switchers, like people trying to move from nonprofits or consulting into tech).

3. Know when to walk away. Not every role is worth your time. If a role feels like a stretch, but you really, really want it, sometimes the better move is finding a bridge role that sets you up for it in 2-3 years.

4. Cold outreach can still work, but it’s hard to do well. Think about how to make your outreach an offer rather than an ask. “Interested in the PM role” will get ignored, but “An idea for your onboarding flow” will get opened.

5. Warm intros help, but the quality matters. A lukewarm referral from someone who barely knows or has never worked with you can be less effective than excellent cold outreach. Know when to use them and how to ask for them. If you do have a connection, send them a self-contained forwardable email that positions you as a strong candidate for them to forward along.

But all of this is easier said than done, right?

So I created 7 custom GPTs based off of our conversation to guide you through each of these steps, using your actual experience and the specific job you want:

Phase 1: Prepare Your Story

1. Job Fit Reality Check: Is this role worth your time?

2. Least Charitable Read Strategist: How might you be misread, and how do you counter it?

Phase 2: Build Your Evidence

3. Proof of Work Builder: Create an artifact of your work that shows how you think

4. Spearfishing Demo Planner: Build a functional prototype in about a day

Phase 3: Make Contact

1. Cold Outreach Subject Line Generator: Write subject lines that get opened

2. Warm Intro Request Coach: Ask for referrals without being awkward

3. Forwardable Email Coach: Write an email your connector can forward with one click

They’re all free to use at writerbuilder.com/jobsearch. Happy searching!

By the way, if you manage people, you have this same opportunity to build simple tools that unblock your team and level up their work. Think about how much of your team’s time is spent waiting on you for feedback, approval, or the answer to a question.

My Supermanagers course teaches you how to fix that. You’ll learn to turn your knowledge into AI tools your team can use on demand, without waiting for your calendar to open up.

Enrollment closes Wednesday, February 4. If you have any questions about whether it’s a fit for you, or if cost is standing in the way of enrolling, please reach out. I’m happy to chat!

xoxo, hils

More Writerbuilder: