Podcast
AI Is Crossing the Frontier of Human Knowledge | Kevin Weil
The a16z Show
- Models Are Crossing The Frontier Of Knowledge
- Modern LLMs can solve open research problems that humans hadn’t solved yet, crossing the frontier of human knowledge.
- Kevin Weil cites ~10–12 open math problems solved recently by GPT-5.2 and Gemini as evidence of emergent capability. (Time 0:04:36)
- Building Prism Brought Weil Back To Coding
- Weil recounts launching Prism, an AI-native environment for scientific writing, and returning to hands-on coding.
- He describes giving Codex tasks and realizing he wasted meeting time when his agent could’ve been working. (Time 0:08:44)
- Use Agents To Multiply Your Productivity
- Be high-agency and run AI agents in parallel so ideas get implemented while you do other work.
- Kevin Weil describes using Codex agents to fix bugs or implement features overnight or during meetings. (Time 0:12:11)
- Closed Loop AI And Robotic Labs For Science
- Accelerating science combines in-silico reasoning with scaled robotic labs in reinforcement-learning loops.
- Weil explains models think, simulate, design experiments, dispatch them to robotic labs, and iterate on real-world results. (Time 0:12:50)
- Design UX Around Model Thinking Time
- When UX involves longer model reasoning, mimic how a human gives intermittent updates rather than immediate chain-of-thought.
- Weil describes showing periodic progress notes during slow reasoning to avoid exposing full chains of thought. (Time 0:18:39)
- Interpret Data By Investigating Anecdotes
- Don’t blindly follow aggregate data; dig into anomalies and anecdotes to understand underlying user segments.
- Weil warns metrics can mask bimodal behaviors and that anecdotes often reveal distinct user needs. (Time 0:21:10)
- Orchestrate Ensembles Instead Of One Giant Prompt
- Use ensembles and orchestration instead of one giant prompt to improve reliability and cost.
- Weil recommends orchestrator models that call smaller specialized models for tasks like customer service or flows. (Time 0:23:02)
- Emergent Capabilities Create Startup Opportunities
- Rapid emergent capabilities produce fertile ground for startups because new model abilities create unknown product opportunities.
- Weil emphasizes OpenAI won’t have all ideas, so founders can discover new use cases as capabilities appear. (Time 0:28:10)
- Start In Enterprise To Capture Early Value
- Target enterprise first because that’s where economically valuable work and paying customers are.
- Weil notes models do costly compute but can deliver clear ROI for businesses, enabling B2B startups to scale fast. (Time 0:30:10)