What the Heck is Graph Engineering?
Summary
This episode covers OpenAI delaying its Astra model due to critical cyber capabilities, ByteDance training a 10-trillion parameter model, and Alibaba’s new revenue-sharing approach for open-source weights. It also notes Anthropic making Auto Mode the default for Claude Code. The main segment provides a primer on “graph engineering,” explaining how it evolves beyond prompt, context, and loop engineering to design complex, multi-agent organizational systems.
Topics discussed
Intro, sponsors, and episode overview
OpenAI delays Atlas model due to cyber risks
ByteDance trains 10 trillion parameter model
Chinese labs bypass chip export controls
Alibaba's open-source revenue sharing model
Claude Code makes Auto Mode the default
Sponsor reads: KPMG, Blitzy, Robots and Pencils, HyperAgent
Introduction to Graph Engineering
Evolution: Prompt and Context Engineering
Harness Engineering explained
Loop Engineering for single agents
Graph Engineering: Multi-agent systems
Org vs Work graphs and conclusion
Listen ad-free on Castria