Best practices are trapped in your head and your AI has no idea
summary
Claim: capture team taste in code so agents start with context not blank slate. Uses VS Code agents, a prototyping repo with agents.md plus file-based design.md and reusable skills (design tokens, codicons) to generate a Kanban agent-orchestration prototype that actually looks like VS Code. Evidence: saved spec.md, agent-run screenshots and iterative fixes, a dogfooding skill, and moving release cadence from monthly to weekly—showing it works and speeds decisions. Repos and a Team Kit marketplace distribute and update the skills.
AI-written summary · about
chapters (8)
- 0:00Prototypes as currency
- 2:18Infusing taste into AI
- 4:25Iterating with AI orchestration
- 7:00Building blocks for AI context
- 12:55Sharing team intelligence
- 15:06Giving agents eyes and hands
- 16:51Critical AI dogfooding
- 19:57The product brain
related talks
16mDon't Build Agents, Build Skills Instead
29mPairing with the machine: How AI Agents are rewriting the rules of software development
31mSpecialize Your Agents with Skills
01h 02mSkills for the age of AI developer tools
27mAgentic Coding with Discipline & Skill: Practical Techniques and Tips
40mHow To Tame Your Agents?
27mAgentic Engineering: Working With AI, Not Just Using It
45mAI agents in practice - beyond the prototype