Why cultivating agency matters more than cultivating skills in the AI era | Max Schoening (Head of Product, Notion)

What It Covers

Max Schoening, Head of Product at Notion, explains why agency—not technical skill—determines who thrives as AI reshapes product building. He covers malleable software, how Notion's designers and PMs now prototype in code, why great products have one tiny superpower, and how the first 10% of any project is now essentially free.

Key Questions Answered

• •Agency over skills: The defining trait separating high performers in AI-era product teams is agency—the belief that the world around you is malleable and changeable. Skills are now accessible via AI models, so the bottleneck shifts entirely to who actually acts. Schoening points to Notion PMs who rewrote their own job descriptions by moving from PRDs into Figma, then directly into code prototypes, without being asked.
• •The "first 10% is free" framework: AI eliminates the startup cost of any project. What previously required weeks of scoping, writing PRDs, and early design can now be prototyped in hours. Schoening frames this as: the first 10% costs nothing, but the last 10% still represents 90% of the real work. Teams should use this to run parallel explorations—send 10 agents down 10 paths simultaneously before committing.
• •Prototype in code to understand the material: Schoening does not care whether designers or PMs ship production code. What matters is that they build in code to understand agent loops, LLM behavior, and the actual medium they are designing for. Static Figma mockups of AI chat interfaces are like "dead fish"—they cannot convey how the system actually feels. Notion built a small, LLM-friendly playground codebase specifically to onboard non-engineers into this practice.
• •Taste is a trainable prediction model: Taste means running a mental simulation to predict whether a specific audience will respond positively to an idea. It is built through high-frequency reps with feedback loops—identical to how machine learning models train via backpropagation. Schoening notes that designers with strong taste consistently maintain side projects where they own the full product, and they continuously expose themselves to other products to calibrate their internal benchmark.
• •Every great product has one tiny superpower: Successful products are not defined by feature breadth but by one exceptionally strong core mechanic. Examples: GitHub's pull request, Heroku's single-line deploy command, Dropbox's menu bar sync icon, Notion's block-and-slash-command system. The most common product failure pattern is the belief that adding one more feature will finally make the product work. It never does—the core must be right first.
• •SaaS is not dying—it is generalizing: The "SaaS apocalypse" narrative is overstated. The "as a service" component—ongoing maintenance, security, specialist expertise, and scale engineering—remains valuable and difficult to self-build. Schoening predicts software will shift back toward general-purpose tools resembling 1990s categories like word processors and spreadsheets, but these will still be cloud-hosted. Most people do not want to maintain their own software stack, just as most people do not want to hunt their own food.