Nov 2, 2026 · 43 min
Figma turns everyday PM work into an AI operating system
How Figma’s PMs actually use AI every day | Ezra Mechaber (PM, Figma)
Figma’s approach shows how AI adoption depends less on isolated prompts than on shared context, repeatable workflows, and carefully bounded access.
- 1Figma’s PM OS packages 31 skills, hooks, context, and workflows into a shared baseline for product execution.
- 2AutoXP treats growth experimentation as an autonomous operating loop, identifying surfaces, running tests, and surfacing promising results.
- 3Jev evaluates writing in real time, while the broader discussion argues that bad AI outputs often reflect poisoned context or unclear instructions.
Don't miss
Ezra demonstrates how Jev evaluates unfinished writing, detects unsupported claims, and links drafts to relevant data as they are written.
The brief
Ezra Mechaber describes Figma’s PM OS, a shared repository designed to give product managers a minimum viable level of execution across neglected operational work.
The system combines memory, organizational context, agent instructions, skills, and hooks; its premise is that strategic AI workflows fail without reliable foundations.
Figma’s 31 skills cover status updates, artifact publishing, coaching, desk booking, and bug bashing, blending tactical execution with interpersonal and generative work.
AutoXP extends the model to growth, autonomously finding experiment surfaces and running tests so valuable traffic does not go unused.
Jev provides a sharper example of embedded AI: it evaluates unfinished writing, flags unsupported claims, and connects drafts to relevant data while people write.
The conversation ends on a practical constraint: when AI performs badly, the cause may be poisoned context, unclear instructions, or excessive access to personal data.
Weekly email
State of the Pod, every Monday
The week's most-discussed names and stories from podcasts, with the moments behind them. Free, five minutes, unsubscribe in one click.