you are not in the race against slop cannons

why humans who care will always have an advantage in the coming AI slopocalypse

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.

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