Cheeky Pint Bret Taylor of Sierra on AI agents, outcome-based pricing, and the OpenAI board
762 snips
Mar 10, 2026 
Bret Taylor, Sierra cofounder and CEO and chair of the OpenAI board, dives into AI agents, why software may charge for outcomes instead of seats, and why companies miss the real gains by optimizing org charts instead of processes. He also gets into boardroom turbulence, hyper-generalists, and a future where smartphones may fade as agents become the interface.
AI Snips
Chapters
Transcript
Episode notes
Why Code Repos Are Ideal Agent Workspaces
- Bret Taylor argues code repositories fit agents unusually well because context, feedback, and history live in text files with tests and compiler errors.
- He thinks messy markdown memory may beat fancier systems because it mixes durable context with random access more like real memory.
AI Coding May Make Documentation The Real Asset
- Bret Taylor thinks coding sessions should produce durable documentation artifacts, not just code, because the intent and customer problem outlast implementation details.
- He imagines agents markdown files pointing into a living architecture directory that captures PRDs, product rationale, and system context.
Why Shared Context Beats Elegant Agent Architectures
- Bret Taylor is skeptical that MCP-style multi-agent architectures are the long-term answer because they hide context inside sub-agents and leave the top agent sounding robotic.
- He thinks stronger agents will look more like OpenClaw's shared markdown memory, where orchestration has enough history and context to act humanely.