Manifest
Anyone can vibe-code an MVP now. The real difficulty comes after — the agent's output doesn't scale, and how to roll out AI-driven across ten engineering departments — almost no one knows yet.
Playbooks all →
Blog all →
Turn off your skills every six months
An agent environment only grows in one direction: rules get added and never removed. The only honest way to know which of them still matter is to run the tasks without them.
AI knows your code. It doesn't know your product.
How we documented a 10-service product so agents could review PRDs — and what it cost. Context, not model size, is the bottleneck for AI-driven development.
Why senior engineers work better with agents
Not magic, not hardskills. The higher the seniority, the better it goes with the agent — because the skill has been years in the making. Not for agents — for people.
The impact of AI agents on development teams
Productivity jumps roughly 2x, but at a cost — cognitive load, decision fatigue, role shifts across developers, QA, leads, juniors, and technical direction. What changes and what to do about it.