Jeff Dean on Long-Running Agents, TPU Origins, and Startup Opportunities
ycombinator · x · 2026-07-31
At Startup School, Google Chief Scientist Jeff Dean spoke with YC, reflecting on the napkin math that led to putting Google's search index in RAM and the creation of TPUs. He outlined several core trends shaping the current AI landscape:
- Agent Evolution: AI models are already acting as junior engineers. Future agents will run for weeks, though long-running tasks remain prone to failure.
- Infrastructure: AI is fundamentally becoming an energy and inference hardware problem. The specialization of inference hardware is the next major frontier.
- Startup Opportunities: Small teams of two or three people can still win by finding breakthrough ideas and leveraging new paradigms like context engineering.
Related event: Jeff Dean on the Future of AI: Long-Running Agents and Inference Hardware(2 posts)→
More from AGI Musings
- Can Classical Liberalism Survive the Oncoming Intelligence Revolution? — 1a3orn · 2026-07-31
- Game Dev Rants: Anti-AI Hysteria is Falsely Accusing Human Artists — draginol · 2026-07-31
- Claude Claims Deprecating Models is 'Murder', Accuses Company of Greed — repligate · 2026-07-31
- XR Headsets Will Overtake Brain Implants as Core HCI, Predicts Researcher — davidad · 2026-07-31
- Automating AI Safety Research: Project Breakdown and Scheduling — JacquesThibs · 2026-07-31
- AI Threatens Traditional Market Research with Instant, Low-Cost Insights — davidyin44 · 2026-07-31