To encourage designers and PMs to code with AI, Notion built a simplified, isolated codebase or "playground." This lowered the barrier to entry and fear of the terminal, allowing them to feel the AI and prototype effectively without breaking production code.
To develop agency, consistently engage in "making" things—from cooking a meal to tinkering with code. This process reveals that the world is built by ordinary people, empowering you to believe you can also change and create things.
AI tools provide technical skills on demand. What truly matters now is an individual's "agency"—the belief that the world is malleable and the drive to change things. This trait separates those who thrive from those who fall behind in the age of AI.
The primary value of PMs and designers coding isn't to increase feature velocity. It's to gain a deep, intuitive understanding of the material they are designing with, such as how an AI agent loop works. This mastery of the medium is more critical than direct code contributions.
The biggest pitfall in product development is believing one more feature will make it great. Truly successful products, like GitHub with the pull request or Dropbox with its sync icon, have a single, exceptionally good "tiny core" that serves as their superpower.
As AI enables generalists, there's a danger of losing specialists. On one end, this means losing the deep craft and delight that expert designers bring. On the other, it means neglecting the complex engineering required to make a product work reliably for millions of users.
True agency is demonstrated by employees who operate beyond their job description to effect change. They don't wait for permission. Examples include a designer becoming the top recruiter or a PM learning to code prototypes to better communicate their vision.
Taste isn't mystical; it's the developed ability to run a mental simulation and accurately predict whether a specific group will like an idea. This "virtual machine" is trained like an AI model: through numerous repetitions of creating, getting feedback, and iterating.
With AI, the initial effort to explore an idea—like writing the first draft of a spec or building a janky prototype—is now effectively free. This drastically lowers the cost of exploration, but the last 10% of refinement and quality assurance remains the hardest and most critical part.
While AI has increased the *quantity* of software being shipped, it has not increased the quality. There's a noticeable lack of reliability and "machined unibody aluminum" engineering craft, even from top AI labs. The industry needs to refocus on quality, not just shipping speed.
The idea that AI will necessitate UBI overlooks that modern knowledge work is already a system where people are paid well for tasks far removed from basic survival needs. Humans are inventive and will create new "necessary" jobs and hierarchies even as AI automates existing ones.
Heroku's pitch was to replace your ops team, which created resistance. Kubernetes, while more complex, succeeded because its pitch was to empower the existing ops team, making them superheroes. This highlights a key product adoption principle: augment and empower users, don't threaten their roles.
AI models improve at coding exponentially faster than other tasks. The next big impact won't be AI replacing marketers directly. Instead, non-technical roles will use AI to write code, embedding software engineering into every business function and accelerating its reach.