June 26, 2026
Multi-model AI coding is useful when disagreement is visible
Claude and GPT disagreement is useful only when it leads to checks, review, and a clear handoff.
Engineering notes
AI coding needs evidence, not theater.
Short technical notes on multi-model coding, agent orchestration, verification, and the gap between a good demo and code you can merge.
June 26, 2026
AI code verification should reject weak evidence
Useful verification separates author and reviewer roles, records checks, and sends weak changes back.
June 26, 2026
Claude vs GPT for coding is a workflow question
Model choice matters, but assignment, disagreement handling, and verification matter more.
July 7, 2026
GPT-5.6 is not a coding-agent strategy
A grounded read on OpenAI's limited preview, model routing, pricing, and what engineering teams should not claim yet.
July 7, 2026
Claude Fable 5 is a long-horizon agent bet, not a merge button
Anthropic is positioning Fable 5 around long-running agent work. The useful question is how teams evaluate it.