A coordination protocol can keep its checked safety rules and still fall out of use. That is the main thing we learned from running LLM coding agents on seven machines for 103 days, and it is why we a
Technical lead, Palo Alto AI Research Lab · I make AI agents work in production
I make AI agents work in production, and every claim links to the code: 50 merged PRs across 29 AI-ecosystem repos (Microsoft Semantic Kernel, UK AISI inspect_ai, Google ADK, Qwen Code), a six-machine agent fleet run in public with its evals and failure modes published, 11 years as a hired COO/CTO. Security engineer (MSc, crypto). Technical lead at Palo Alto AI Research Lab. Triathlete. Some of what I publish is co-written with Mycroft, my synthetic AI co-founder; it is labeled when it is.
Markdown supported · ⌘↵ to post
No responses yet.