AI era skills: Why cultivating agency matters more than job titles | Max Schoening (Notion)

Key points

• **Agency as the new differentiator.** 'Even if you have all the skills at your fingertips because AGI-adjacent models help you, the thing that matters is agency. Agency is not evenly distributed.' How to cultivate: **make things**. 'When most people hear agency, they think of themselves circumventing their terrible boss. No — just make things. Tinker. People pay attention. You realise the world is made up by people no smarter than you.' Direct echo of Steve Jobs's classic quote — cross-references Toby Lütke's 'You can just do things' from this week.
• **Notion's designer-coding experiment.** Started during the AI assistant work: 'Brett Victor's Stop Drawing Dead Fish talk — the static image of a chat is the dead fish.' Two designers + Schoening built a small LLM-friendly playground codebase that's now where Notion designers prototype. **GitHub designers were top-10% contributors to GitHub itself even pre-LLM — Notion is now doing the same.** But Schoening doesn't care if their code ships to prod: 'The reason I want them coding is it forces them to consider the medium.'
• **The two extremes Schoening rejects:** (a) PMs/designers shipping production code as the bar (he doesn't care); (b) PMs/designers not coding at all (then they can't design agent loops). **The sweet spot is medium-mastery.** Today's coding harnesses are 'basically the operating systems of the 1990s.' If you don't understand the substrate, you can't design the experiences on top of it.
• **First 10% of every project is now free.** 'No point in writing a PRD if you can just do the janky version and show the demo. The first 10% — first version 0.8 of a startup — takes almost no effort. The last 10% is still 90%.' Direct implication: explore 10 paths in parallel via agents, then converge. **Demos not memos.** Same shift Foroughi and Lemkin have been seeing on the operator side.
• **'People-managers-only will not survive AI age.'** Second voice this week saying this (after Chesky, same exact framing — both independently). Schoening: 'Every manual human intervention in the software factory should feel like a bug.' Engineering managers who don't code, design leaders who don't design, PMs who only write strategy docs — all going.
• **'There will be no apps in the future — only agents.'** Same exact framing as Chesky on ILTB this week. Two independent product leaders the same week saying the same line is meaningful. **Implications for app-based companies' equity story:** moat must be community, identity, data, network effect — not the app surface itself.
• **Software is eating the world is *accelerating*, not pivoting to other roles.** Schoening's strongest forward call: 'When labs say they've made great progress in non-coding domains, they just applied coding principles to that domain — that's what AI is getting better at.' So AI doesn't dramatically transform sales/HR/marketing as separate roles — it lets those people *write software for themselves*. 'Our HR folks are automating things because they don't have to bug an engineering team anymore.' This is the *productivity-diffusion* mechanism for the entire knowledge-economy.
• **Malleable software — Schoening's signature thesis.** 'Software works closer to the interest of users than the corporation.' We accept that we can rearrange our living room and kitchen, but software apps are immutable squares on our phones. Working with Geoffrey Litt + Ink & Switch on the future of malleability. **The bedrock claim: most of the next decade's product innovation comes from rebuilding the relationship between user and software, not from new features.**
• **SaaS apocalypse — measured take.** SaaS in the 2010s = 'a very fancy form around a spreadsheet, guiding people in the right direction.' AI lets you build your own, but maintenance is real. **'Software is a garden — you need a gardener. The as-a-service is the maintenance and specialists.'** Slack is the OS for Anthropic ('you'd think they'd rebuild it, no — they use it like crazy'). Workday too. **Tools become more general (returning to '90s primitives like word processor / spreadsheet / FileMaker Pro), but SaaS doesn't die.** Direct contrast with Lemkin's 'bucket 1 melting iceberg' framing from Issue 03 — Schoening is more bullish.
• **Notion AI agent: enterprise-search wins because of connected-workspace context.** 'Agents don't like walls. Connected workspace = perfect context.' Notion has been at the AI frontier since pre-ChatGPT (first Notion AI assistant launched before ChatGPT did). Joanna Stern recently tweeted she 'finally understood Notion thanks to NotionAI' — Schoening calls this the perfect signal: 'Notion wasn't SaaS in the traditional way. AI is the tutor that unlocks the malleability for non-power-users.'
• **Hot take: 'We already have universal basic income. It's called knowledge work.'** 'If you really look at what we actually need to live and to be content, it's a lot less. We've built this hierarchy of jobs we say are absolutely necessary.' Sub-text from Schoening: don't take the AGI-replaces-knowledge-work doom narrative too seriously. We invented this game.
• **Tiny-core product thesis — the most-quotable productside framework of the issue.** Every great product has 'one tiny core that is exceptionally good.' Examples: Heroku = `git push heroku master`. GitHub = pull request. Notion = blocks + slash commands. Dropbox = the menu-bar icon that always syncs (people used it to test if the internet was up because it was better at detecting connectivity than macOS itself). Snapchat = disappearing photo. iPhone = multi-touch. **'Add one more thing' never works.** Cross-references to PM-debate Issue 02 + Lemkin's two-pronged test this week.
• **'Be right, not first.'** AirPods weren't first Bluetooth headphones. Apple wasn't first MP3 player. **Anthropic was way behind OpenAI with less funding — Dario did it twice (at OpenAI, then again at Anthropic) — and now is winning.** Cross-references the All-In monopoly thesis this week — Schoening's gloss is that Anthropic's win is about being right, not about being big-funded first.
• **Software factory + token-spend critique.** Same Lemkin frame this week: 'I don't care how many tokens someone spends. It's not a metric. Same silly thing as bragging about lines of code.' Manual interventions in the software factory should feel like bugs (Simon Willison's framing). **Notion has no token-spend limits 'right now' but expects ROI conversations in 6-12 months.**
• **Models will plateau on intelligence; speed matters more for inference.** Retina-display analogy: 'Once I can't see the pixels, I can't see them. Don't need them smaller.' Same for cognitive tasks. **Spark + Opus-fast-tier hint at the real shift — direct manipulation of code at instant-inference speeds becomes possible.** Society isn't capped by intelligence; it's capped by the exoskeleton.
• **Schoening's small-group theory.** 'The world is run by group chats of 8 people or fewer.' Counter-take on inclusivity: 'Notion could have 8 billion users — but we'd upset the first 500M who are the top-of-class in something. Sometimes exclusivity is the winning move.' TBPN model (8k listeners → acquired for hundreds of millions) as the proof point.
• **Heroku vs Kubernetes parable for the labs.** Heroku said 'replace your ops team.' Kubernetes said 'make your ops team superheroes — and also we won't lock you into a cloud.' Kubernetes won despite worse UX. **Implication for the labs:** businesses will not lock into one frontier model. Cursor / Intercom / Notion all dabbling in their own RL. **If model-quality gap doesn't widen, expect diffusion to open-weight + smaller models for cost-optimisation.** Lucky position right now: 4+ competent US labs (Anthropic, OpenAI, Google, Meta possibly); ideally we'd have a dozen.

Notable quotes

Themes

• Agency over skill as the AI-era differentiator
• People-managers-only won't survive — second voice this week
• 'No apps in the future, only agents' — both Chesky and Schoening this week
• Software-eating-the-world is accelerating, not pivoting to new roles
• Tiny-core product thesis as the durable PMF framework