Notion's Head of Product: AI Killed the Skill Barrier, But Most People Don't Know What to Do With Superpowers

Source: Lenny's Podcast | Published: 2026-05-02T15:01:17Z

Max Schoening argues AI makes the first 10% of any project free — but even with 90% done, the last 10% is still the real 90%. Exploration is cheap; shipping to 100 million users is not.

The First 10% Is Now Free

Here's how Max describes the shift in his workflow: the first 10% of every project now costs zero effort. Write a PRD? Just build a rough demo instead. Explore ten paths? Send ten agents down each one and see which pans out.

He pushes this logic to its extreme — a startup's first version can be stood up at virtually no cost. But he immediately adds: even if you're generous and say the first 90% is done, the remaining 10% is still the real 90%. Exploration is cheap. Making something work for a hundred million people is still expensive.

During his GitHub days, their product reviews had a mantra: demos not memos. Now "give me something I can react to" has become dramatically easier. Waterfall development is pointless in this context.

Agency Is Not Evenly Distributed

Max has noticed a pattern: once AI removes the skill barrier, what differentiates people is no longer "can you do this" but "will you go do this." He puts it bluntly — you used to always have the excuse of "I could never do this because I don't know X." That excuse is gone now.

Inside Notion, he uses a phrase to fire up himself and his team: Drive Notion like it's stolen. The idea: you're not the founder, the company already has incredible product-market fit, but can you act like you stole this car and floor it?

He gives two examples. One is Brian Lovin, who blurred the lines between engineering and design while also taking on recruiting — "what kind of people does this org need? I'll go find them." The other is product manager Eric Lou, who one day asked Max: "If you started a company, would I be in your first ten hires?" Max said no — I wouldn't need a product manager. Eric's response: then I need to become someone you'd hire in your first five. So he went from writing strategy docs to designing in Figma to writing prototype code himself.

Designers in the Terminal

Notion's designers now use terminal tools extensively. It started when Max joined and found the team designing chat interfaces in Figma — an interaction that fundamentally needs to be felt, reduced to static mockups. He references Bret Victor's talk "Stop Drawing Dead Fish": a static mockup of a chat interface is exactly that dead fish.

So he and two designers built a minimal code playground — intentionally making the codebase LLM-friendly, using toolchains that LLMs excel at. The goal was to get designers past their "fear of the terminal," after which everything becomes a conversation.

The key detail: Max doesn't care whether designers actually push code to production. What he cares about is designers thinking in the medium of code. If you don't understand how an agent loop works, you can't design an agent product well — and right now, the only way to understand an agent loop is to build one in code.

"I'd rather have a designer who deeply understands how agent loops work than one who can only tweak styles."

Software Quality Is Disappearing

Max has a contrarian take on the current "vibe coding" wave: over the past 12 months, the quantity of software may have increased, but the quality of software hasn't noticeably improved. Truly reliable software is still hard to find.

He draws an analogy to hardware. If you're doing a hardware startup, your first version is 3D-printed with visible layer lines — nobody would mistake it for a consumer-ready product. But in the software world, people seem to have forgotten the long engineering road from prototype to production — how do you make this thing run reliably in a hundred million people's hands? How do you optimize your factory's yield rate?

He doesn't spare the AI labs either:

"I love these tools, I live in the CLI. But every two weeks there's a regression, issues fixed three weeks ago resurface, and the TUI frame rate is embarrassingly low."

What he wants to see is "Apple-level unibody aluminum engineering" — the entire industry needs to rediscover its obsession with software quality.

Taste Is a Virtual Machine Running in Your Head

On "taste" — one of the AI era's most overused words — Max offers a delightfully geeky definition: taste is a virtual machine running in your head. Given an idea, you can predict whether your target audience will love it.

How do you train this VM? Same way you train a model — input an idea, observe the reaction, backpropagate, repeat. No shortcuts, just reps. He points to Japanese craftsmen: someone who's been painting bowls for decades has taste as a function of time.

He's observed that designers with great taste share two traits: they always have a side project where they own the full stack, and they constantly try new tools — they're the slightly annoying person on the team who's always recommending something new.

Interestingly, he's not convinced taste is humanity's last moat. If taste is fundamentally an "input → feedback → adjust" loop, that looks suspiciously identical to how models are trained.

The Secret of Great Products: A Tiny, Powerful Core

Looking back at the products he's worked on, Max distills a pattern: every great product has a tiny, supremely powerful core.

The iPhone's is multi-touch. GitHub's is the Pull Request. Heroku's is `git push heroku master` — one command, and local code becomes a live URL. Dropbox's is that sync icon in the menu bar — it was better than macOS itself at knowing whether you had internet. Notion's is blocks and the slash command.

The counter-lesson comes from his own experience. In 2014, he founded a document tool and obsessed over the editing experience — markdown folding, features Obsidian wouldn't ship for years. That same week, Notion pivoted from website builder to document collaboration, and True Ventures pulled his term sheet because of the overlap.

Notion's early editor was terrible — everything was blocks, and you couldn't even select text across them. But it didn't matter. The core was right; everything else could be iterated. His own product missed the core, and no amount of feature additions could save it.

"If you're stuck in a loop of 'one more feature and the product will be great,' it never will be."

The SaaS Apocalypse Is Wildly Overstated

Faced with the "AI will kill all SaaS" narrative, Max's take is: the "as a service" part is what's actually valuable.

He tried rebuilding Notion over a weekend. The conclusion: nobody actually wants to maintain an entire software stack. Software is a garden that needs constant tending. What you're paying for is a team of experts continuously thinking about one problem.

He uses Anthropic as an example — the whole company is deeply reliant on Slack. If any company has the capability to build their own replacement, it's them. But they don't, because the ROI of spending that time building AGI is higher. The decision tree for just "sending a notification" in Slack is so complex that only decades of accumulated knowledge gets it right.

His prediction: software will return to the '90s model of general-purpose tools — word processors, spreadsheets, FileMaker Pro — but still delivered as SaaS.

Malleable Software: Your Digital Life Shouldn't Feel Like a Locked-Down Apartment

Max has long championed "malleable software" — software should serve users' interests, not the interests of the company that built it.

His analogy: imagine living in an apartment where you can't rearrange the living room furniture and the kitchen layout is dictated by someone else. Nobody would accept that. But that's exactly the state of software today — every app is a sealed box, with interface, data, and behavior all glued together. You can't change even a small detail.

The other extreme is running your own Linux distro — total freedom, but you also have other things to do and don't want to wrestle with trackpad drivers every time.

AI is creating an opening. People are starting to build their own tools — a form of malleable software. But Max believes it has to be built on a platform that encourages this behavior. Otherwise you just end up with everyone having their own isolated little tool, with no collaboration to speak of.

Model Intelligence May Hit a "Retina Display" Moment

Max offers an intriguing analogy: after Retina displays, pixels could keep getting smaller and I wouldn't notice. Is there a similar saturation point for cognitive tasks?

His intuition: for most knowledge work, model intelligence becomes "good enough" past a certain threshold. At that point, the more important optimization axes are speed, local execution, and cost — not more intelligence.

This leads to another implication: if inference speed approaches real-time, the human-computer interaction model might return to direct manipulation — shaping code like clay in real time — rather than today's async pattern of "queue up a bunch of tasks, go for a walk, come back and review the results."

"Labs assume people always want the smartest model. But we don't staff every role at a company with a PhD either."

Don't Flex Your Token Consumption

Max claims to be Notion's highest individual token consumer as a PM (possibly second only to automated security scans). The company's current policy is no cap — he thinks optimizing this metric right now would be a mistake.

But he's noticed a worrying trend: people are starting to flex their token consumption at each other, the same way they used to brag about lines of code written per day. Meta's recent leaderboard is a case in point.

He understands Meta's motivation — in an organization of tens of thousands, you do need some push to get people out of their comfort zones and start delegating outer-loop work to agents. But as a metric, token consumption is just as meaningless as lines of code.

His prediction: within 6–12 months, many companies will start seriously discussing the ROI of AI spending. That's going to be an uncomfortable conversation.

Knowledge Work Itself Is UBI

Max has a half-joking, half-serious hot take: we already have universal basic income. It's called knowledge work.

His logic: if you honestly examine what humans truly need to sustain life and find fulfillment, it's far less than we think. We've built an entire hierarchy of "absolutely essential" positions. Sitting in air conditioning, placing the right letters in the right order on a computer, and collecting a high salary — that's already a pretty luxurious arrangement.

As for what happens when AI replaces jobs? He says humans will do what they've always done — invent new reasons to insert themselves into the loop. The form may change, but our creativity in that department has never disappointed.

Don't Let FOMO Decide How You Spend Your Heartbeats

Max has an observation about young people in Silicon Valley: it's now crowded with people who don't genuinely love computers. They're making choices driven by narratives about "the last train" and a "permanent underclass."

His advice isn't "don't work hard" — on the contrary, he thinks 18 to 25 is the time to work relentlessly. But he wants people to dial down the amplitude of their anxiety, and then go read history. History repeats itself far more often than it does anything entirely new.

If you have agency, if you're not fixated on demanding certainty about how the world will unfold, you'll probably be fine.