I heard recently that US employment numbers went up.

US (as of June)

  • Nonfarm payrolls: +57,000 (a slowdown—roughly half of what the market expected)

  • Unemployment rate: 4.2% (down 0.1 percentage points from the previous month)

With AI moving as fast as it is, everyone and their dog is out there vibe coding, shipping real services, cranking them out, hustling to promote them. I'm no exception. Right now I'm juggling a personal project, a slice of a company project, and some freelance work on the side.

(For context: my background is IT product—so, PM work. I've done design too, and I'm doing some of the development myself.)

I wanted to jot down a few things I've been noticing along the way.

Thanks to AI, getting an idea out of your head and into the world no longer requires knowing how to code. For me personally, it's gotten to the point where I can move almost everything in my head straight onto the screen, more or less intact.

But here's the thing—when it's just me, this never causes problems.

Working solo, maybe my thinking is shallow, maybe the model's having an off day, maybe I'm charging ahead on something I don't actually understand conceptually. Doesn't matter. You wrestle with it, roll it around a few times, and eventually the thing gets built.

The trouble starts the moment this "scales" into a team or an organization. And weirdly, the exact problems I used to hit at my old company float right back up to the surface—except now they're even easier to see. Ha.

Let me give you an example.

Say a founder has a service in their head. They build the whole thing themselves. The founder's happy—what they wanted is finally sitting right there in front of them, visible.

Then talented people start joining, one by one. And the instant they do, everyone's competing interests kick in and the original vision starts drifting.

The founder looks at Feature A and thinks B. The team members think C, D, E, F… each one reading it through their own experience. Feature A gets translated into a handful of different languages, AI joins the chorus, and out pops some marvelous thing called "AANDHSMDSADN."

Don't get me wrong. I'm not saying these weird mutant outputs are bad. (Honestly, hallucinations and messy processes like this are often exactly where the most distinctive, one-of-a-kind ideas come from!)

The problem isn't the output. It's personal taste, interest, and understanding.

When you decide Feature A is what you need, but your interest in it and your grasp of it—setting AI aside, the actual synapses in your brain—aren't really holding onto A, then your teammates start building from C, D, F… each on their own version of understanding. And all the while, AI is listening to every word of the conversation, which just sends things further up the creek.

This feels like the "curve of despair" a lot of startups are riding right now. Cognitive debt, piling up while everyone keeps deferring the work of actually understanding. AI, which was supposed to fill that gap and hand you the perfect result. And the result? A row of red X's.

Eventually everyone, founder to teammate, fills up with the same feeling: nobody understood what I meant. And that feeling quietly hardens into an assumption—this product probably isn't worth killing myself over—which becomes giving up.

"I see it this way, so why are they like that? Just do this much, that's all it takes—why not?" Once that label gets slapped on, it stretches into something bigger: no matter how hard I try, I can't change that person—so why should I bother trying at all?

I'm writing this partly because it's what I'm actually feeling, and partly because I keep circling back to one question: how should organizations change from here?

The feeling of having read something isn't the same as understanding it

You can structure a document beautifully, and the act of reading still won't guarantee understanding. It's easy to trace sentences with your eyes and walk away convinced you've remembered or understood something you haven't.

This idea comes from Andy Matuschak's "Books Don't Work," and from Quantum Country by Andy Matuschak and Michael Nielsen. Quantum Country embeds spaced-repetition quizzes right inside the text, so the reader has to actively recall what they just read.

I think you could apply the same trick to code documentation. Something like dropping five interactive questions about the changes at the very end.

Debugging for someone vs. building them a way to understand

Here's another one. "The agent debugs it for you" and "the agent builds you a tool for understanding" lead to completely different outcomes.

"The agent finds and fixes the problem" and "the agent builds a tool the human can use to explore the problem" are different in kind. The first gets you a fast result, but your understanding of what's happening under the hood might barely move. The second gives you an actual exploration—you inspect the execution state, form a hypothesis, check the result.

Hand part of the coding to the agent, sure—but keep the core thinking and exploration for yourself. It's about expanding the agent's role from answer-provider to maker of environments for understanding.

A team has to understand together

image.png

It's all very theoretical, but the core is simple: a team has to understand together.

One person understanding the system isn't enough for teamwork. Everyone needs to hold the same mental model, so that even a short phrase conjures the same structure and behavior for all of them. When you share a vocabulary and a set of images, it's easy to toss ideas back and forth on the fly and grow them in new directions together.

Do the opposite—everyone isolated with their own separate agent—and you get: the same terms used to mean different things, technical plans whose premises nobody shares, outputs that combine while understanding never does.

A team's use of AI isn't just a question of boosting each individual's productivity. It's a question of building sharedunderstanding across the whole team.

This process—brains wiring together alongside a team might just be the most powerful thing a team can have going forward.