what's the point of managers?

spoiler: the hard part was never the work itself

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.

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