Viewpoint
Questions on agentic systems with markdown and cron jobs
Questions on agentic systems with markdown and cron jobs
Garry Tan asks about the mechanics of agentic systems, mentioning markdown files, cron jobs, and pointing agents at sufficient data to enable out-of-distribution discoveries.
- Speaker
- Garry Tan
- Topic
- Agentic Systems
- Source timestamp
- 28:45
- Context timestamp
- 28:15
More from this interview
- Wang's path from competitions to Scale AI idea
- Working at a company teaches valuable operational insights for founders.
- Scale AI pivoted from medical agent after feedback
- Training data gap led to Scale AI's founding
- Investor perception of Scale AI shifted from skepticism to enthusiasm.
- Companies need uncommon convictions early
- Founders improve through continuous learning
- AI bottleneck is adoption, not model progress
- AI agents let startups outcompete incumbents
- Meta envisions decentralized personal superintelligence
- Meta targets billions of AI-driven businesses
- Wang rebuilt Meta's lab, launched Muse Spark models
- Frontier AI needs a scientific operating model with ecosystem growth.
- Meta pursues decentralized AI with open-source models and a developer harness.
- Cheap Meta models aim to democratize access across growing AI waves.
- Superintelligence timing debates waste time as AI progress is inevitable.
- Abundant intelligence will make vision and ambition the scarce resources.
- Builders must help governments and enterprises adapt to AI risks.
- Systems thinking remains crucial as orchestration replaces direct coding.
- Civilization needs a deeper philosophical compass as change accelerates.
- Agentic looping offers astronomical opportunity by optimizing business outcomes.
- Building agentic systems is mundane mechanics, not magic.
- Advice for younger self: keep conviction and find exponential trends
- Scale AI's success from first-principles thinking
- Lucky time to start a company with AI tools