Playbook
Tasteful Skills
“Tasteful Skills” argues that the best agent skills are not documentation or best-practice lists.
Back to Podcast Digest
AI era skills: Why cultivating agency matters more than job titles | Max Schoening (Notion)
TL;DR
Agency is becoming the real moat — Max Schoening argues that with AI putting skills “at your fingertips,” the people who win are the ones who treat the world as malleable, not the ones clinging to rigid job definitions like PM, designer, or engineer.
At Notion, coding matters less as output than as a way to think in the right material — Schoening pushed designers and PMs into LLM-friendly playgrounds and terminals not so they could all ship production code, but so they could actually understand agent loops, software constraints, and the medium they’re designing for.
AI has made the first 10% of a project basically free — instead of writing PRDs, teams can now build a rough version, send “10 agents” down 10 paths, and give others something concrete to react to, which compresses exploration and kills a lot of waterfall planning.
Great products still hinge on one tiny superpower — his test is whether a product has a small core that’s exceptionally good, like GitHub pull requests, Heroku’s “git push,” Dropbox syncing, or Notion blocks and slash commands; adding “one more feature” is usually the death spiral.
The SaaS apocalypse is overhyped because people don’t want to maintain their own software gardens — he thinks software will get more general and more malleable, but most users still want the Costco version, not to rebuild Notion or Slack themselves and maintain it forever.
Taste is basically a prediction engine trained by reps — Schoening defines taste as being able to simulate whether a specific in-group will love something, and says you build it the same way models do: repeated making, feedback, iteration, and lots of exposure to great work.
The Breakdown
Why Notion got designers and PMs into code
Schoening opens with the origin story: Notion was designing AI chat interfaces in Figma, but static mockups felt like Brett Victor’s “dead fish.” So he and two designers built a scrappy, LLM-friendly playground where people could prototype in code without touching the scary main codebase. The point wasn’t “everyone must ship production code” — it was getting people onto the treadmill so they could feel the medium.
Coding isn’t the goal — understanding agent loops is
He’s blunt that he doesn’t actually care whether designers’ code lands in production. What matters is that coding forces product people to interrogate the material they’re working with, especially now that agent loops are becoming central to software. He’d rather have a PM who deeply understands how agent systems behave than one who can only tweak UI styles with a coding assistant.
The big skill shift: from credentials to agency
The most important trait in this era, he says, is agency — the belief that the world is made of things you can change. He gives Notion examples like Brian Lovin “driving Notion like it’s stolen,” recruiting talent outside his formal role, and Eric Liu reshaping himself from strategy-doc PM into someone who prototypes directly. Schoening’s advice is refreshingly concrete: start making things, because tinkering is what teaches you that reality is editable.
Malleable software, useful design, and the danger of losing specialists
His long-running idea of “malleable software” is that software should serve users’ interests more than the corporation’s, more like rearranging your living room than living in a locked-down app world. But he’s also wary of the current chaos: if roles blur too much, we may lose specialist craft in design and the kind of engineering that makes software reliable for 100 million people. He uses physical metaphors throughout — 3D-printed prototypes versus factory-grade manufacturing — to say today’s discourse is too focused on tokens and features, not enough on durable quality.
Why SaaS isn’t dead, even if software gets more customizable
On the “SaaS apocalypse,” his take is basically: yes, old SaaS changes, but no, people do not want to maintain the whole stack themselves. He compares rebuilding software to hunting your own steak instead of buying it at Costco, and says the “as a service” part still matters because software is a garden that needs constant tending. His bet is that tools get more general, more AI-assisted, and more malleable — not that everyone suddenly becomes their own IT department.
AI changed product work by making exploration cheap
The cleanest line in the whole conversation: “the first 10% of every project are now free.” Instead of long PRDs, teams can spin up rough prototypes instantly and give collaborators something real to react to, which makes “demos not memos” much easier to live by. He thinks this increases shots on goal dramatically, but also warns that the last 10% — quality, reliability, reconciliation — is still the hard part.
Faster models may matter more than smarter ones
Schoening is unusually skeptical that ever-smarter frontier models are the whole story. For many knowledge-work tasks, he thinks we may hit “good enough,” after which speed, cost, local execution, and UX matter more — like a retina display after you can no longer see the pixels. That’s why he’s more interested in AI as an exoskeleton that augments people than as a god in a box doing everything for them.
Taste, tiny cores, and his weirdly sharp closing philosophy
Asked how to build taste, he gives a nerdy but memorable definition: taste is running a virtual machine in your head that predicts how a specific in-group will react. You get there with reps, side projects, and exposure to great objects — hence Notion conference rooms named after iconic products like the Macintosh and Porsche 911. He closes on a bigger note: don’t let AI panic make you live frenetically; go for a walk, notice that the world was built by people no smarter than you, and act accordingly.
Share
Keep Reading
Two Ways to Follow
The Weekly Echo. The inbox-shaped summary of what mattered.
New editorials announced here.
