Developer argues GitHub is too deterministic for probabilistic AI coding workflows
fforres · x · 2026-10-17
fforres argues GitHub is rightfully deterministic but needs to become friendlier to today's probabilistic AI agent workflows. Suggestions include simpler merge queues/priority lanes, local signing to skip redundant checks, checks-as-prompts, intent-based conflict resolution, MCP events as first-class citizens, simpler identity management with short-lived JWTs, and no per-seat pricing. Their company has explored these ideas (gateway/merger/events over Cloudflare Artifacts) with good results. Core thesis: Git handles history, conflicts suck, and GitHub shouldn't be a blocker to shipping.
More from coding & agent
- OpenAI launches community site for Codex-built hardware projects, including a Purble Palace port to ModRetro Chromatic — OpenAIDevs · 2026-10-17
- Claude engineer runs 5 Claude Code agents in parallel terminals — now routine — adonis_singh · 2026-10-17
- AI auditing AI: who checks the checker? — CrashTestingAI · 2026-10-17
- Using Claude Code is still 'VERY frustrating', says founder amid hype over superhuman coding claims — robleclerc · 2026-10-17
- AI agents are having their Kubernetes moment: agent logic sinks into the platform layer — bibryam · 2026-10-17
- Engineer's long-form guide: building your own personal software factory — maddiehfaulkner · 2026-10-17