Steve Yegge. Long-time software engineer who spent years at Amazon and Google before going independent. He is building Gas Town and Wheelhouse, his own agentic coding tools, and writing from primary experience about what it looks like to run hundreds of AI coding agents in parallel. His essays on the future shape of agentic engineering are among the few accounts from someone who has shipped these systems at scale rather than theorized about them.
Operating themes
- Harness-native architecture Building agent harnesses as a first-class part of the application, not an external wrapper, to survive model upgrades and behavioral drift.
- CI/CD at agent scale How traditional merge queues and deployment pipelines fail when agent count moves from single digits to hundreds, and what must replace them.
- Model welfare as an engineering discipline Treating agent identity and continuity as practical engineering concerns that affect output quality and debuggability.