Agency Beats Skills in the AI Era — Here's What That Actually Means for Builders

Lenny's Newsletter just dropped an interview with Max Schoening — Head of Product at Notion, formerly VP of Design at GitHub, ex-Heroku, ex-Google, two-time founder — that contains one of the most useful framings I've heard for what's actually changing in product work right now.

Schoening's argument: in the AI era, agency matters far more than skills. The first 10% of any project is now effectively free, and the people who'll win from here are the ones who actually go do something with that.

Let me unpack what he's saying, because the title sounds like a LinkedIn humblebrag but the actual content is sharper than that.

What "agency" actually means here

When Schoening uses the word agency, he's not talking about the trendy "AI agent" definition. He's talking about it the old-fashioned way: the willingness to take initiative, make decisions, and drive change without permission.

That trait used to be table stakes for senior people and rare in juniors, because juniors didn't yet have the technical skill to back up their ambition. The skills bottleneck enforced humility — you had to spend years learning the craft before anyone trusted you to run with it.

AI just collapsed that bottleneck. A designer can prototype a working app in an afternoon. A PM can write functional code without a developer. An engineer can ship in three languages they don't actually know. The skill floor for taking action is no longer the gating factor.

Which means, as Schoening points out, agency is now the gating factor. The people moving the fastest aren't the ones with the deepest expertise — they're the ones willing to push the button.

"The first 10% is free"

This line from the interview is the one that should be on every product team's whiteboard.

For 20 years, the painful part of any new project was the start. Spinning up the repo. Designing the schema. Doing the wireframes. Writing the boilerplate. The first 10% of the work was where most projects died — not because the idea was bad, but because the activation energy was too high.

AI tooling has made the first 10% effectively free. You can prototype, scaffold, design, and demo something faster than it used to take to schedule the kickoff meeting.

That changes the calculus on every "should we even try this?" debate inside a company. When the cost of the first 10% was a quarter of design and engineering time, you needed strong conviction to start. When the cost is a Saturday afternoon, you can just go see.

The organizations that internalize this are going to ship 10× more experiments than the ones who don't. The ones who don't are going to wonder why their roadmap looks identical to last year's.

The "tiny core" theory of great products

Schoening's other big idea in the interview is what he calls the tiny core theory of great products. The argument is that the products that have lasted have one elegantly simple foundational mechanic that everything else hangs off:

• The iPhone had multitouch.
• GitHub had the pull request.
• Notion has the block.
• Dropbox had the menu-bar icon.

In each case, the rest of the product is built on top of one beautifully designed primitive. Take the primitive away and the rest collapses. Get the primitive right and the rest has somewhere to grow into.

This matters now because the AI era is going to flood the market with products that don't have a tiny core — wrappers around models, dashboards on top of APIs, generic agent platforms that do everything and nothing. The products that survive 24 months will be the ones that found a primitive worth building on top of.

If you're building right now and your product doesn't have a one-sentence answer to "what's your tiny core?" — you're probably building a feature, not a product.

The quality-quantity gap is real

The interview also makes a point that's been quietly true for a year and finally has a name: there's a gap between AI's quantity output and AI's quality output, and that gap is where opportunity lives.

Volume of software shipped per developer is way up. Quality of software shipped per developer hasn't kept pace. AI lets one person produce 10× the output of a year ago — but the median output is also 10× more average than it was a year ago, because everyone is using the same defaults.

The builders who are actually distinguishing themselves are the ones using AI to increase both speed and taste — not just speed. That means time spent on:

• Refining defaults instead of accepting them.
• Looking at output and saying "this is wrong" instead of "this is fine."
• Holding standards for visual polish, microcopy, and edge cases that AI doesn't volunteer.

The quality-quantity gap is the same gap that's existed in every creative profession forever — but it's wider now, faster than it's ever been. Builders who treat AI as a speed multiplier without a taste discipline are about to find out their work is interchangeable. Builders who treat AI as a taste multiplier — by spending all the time saved on quality — are going to look like they came from a different planet.

"Drive it like it's stolen"

The phrase Schoening uses for Notion's shipping philosophy is great: drive it like it's stolen. Aggressive, decisive, no perfectionism, no endless iteration before shipping. Make a decision, push it live, learn, adjust.

The old defense of perfectionism was that shipping bad work was expensive — your team would have to maintain it, debug it, refactor it. AI has changed that math too. Refactoring, rewriting, and replacing code is significantly cheaper than it was. The cost of being wrong has gone down. The cost of being slow has not.

Which means: ship sooner, decide faster, edit harder. The companies that act like every decision is reversible are going to outrun the ones still treating product decisions like they're permanent.

The SaaSpocalypse take is right

The last big point Schoening makes — and I love this — is that the "SaaS is dead because of AI" narrative is way overstated. AI is going to eat some SaaS, sure. But existing companies have distribution, data, customer relationships, integrations, and compliance posture that a wrapper-on-an-API simply doesn't.

If you're a SaaS founder right now, your real risk is not adopting AI fast enough internally, not getting replaced by an AI-native competitor. The replacement story is mostly noise. The complacency story is the actual threat.

The Bottom Line

Max Schoening's interview is a tight, useful read for anyone building product right now. The thesis is simple: AI eliminated the skill bottleneck, which means the differentiator is now agency — the willingness to start, ship, edit, and decide. Pair that with a tiny-core product, a discipline around quality over volume, and a "drive it like it's stolen" approach to shipping, and you're already operating better than 80% of teams. The skills floor is gone. What's left is whether you actually do the thing.

Note: This article was written in editorial voice — mads-brain knowledge base was unavailable during processing.