Y Combinator Startup Podcast Tokenmaxxing: How Top Builders Use AI To Do The Work Of 400 Engineers
274 snips
May 8, 2026 
Gary Tan, YC partner turned builder who returned to coding and ships agent-driven tools like GStack and G-Brain. He describes rebuilding products quickly with AI coding agents. He talks about token-maxxing, agentic workflows like conductor/queue/QA, and why personal, user-controlled AI will reshape how one person builds powerful software.
AI Snips
Chapters
Transcript
Episode notes
Structure Agent Workflows With Plan And Eng Skills
- Use prompting skills (plan, eng, CEO) to structure work: outline architecture, tests, UX, and edge cases before coding.
- Gary queues plans in Conductor, approves them, and lets Claude implement with automated tests and CI steps.
Teaching Agents To Run Browser QA
- Gary automated QA friction by integrating Playwright and browser-based QA into his agent pipeline.
- He queued 15 features that passed unit/integration tests but still required manual QA, so he taught agents to run UI checks instead.
Keep Humans In The Loop For Agency And Taste
- Always keep a human in the loop: use 'ask user' questions in plans to surface domain knowledge and ensure the agent aligns with the user's intent.
- Gary cautions against fully autonomous software; humans supply agency, taste, and priorities.