Stop keeping agent sessions local: shared traces enable self-improving agents
vikvang1 · x · 2026-10-23
Dev Bertie Holmes argues agent conversations shouldn't live on local machines but in a shared store. That unlocks scheduling self-improvement agents to mine past runs for failure modes, and replaying old conversations against different models for benchmarking. His company Warp Factories runs agents in cloud sandboxes with traces available via API, plus an inbox view for conversations needing attention.
More from coding & agent
- What is a "head of Evals"? The invented job title, explained — danshipper · 2026-10-23
- Figma launches Code Layers, bringing interactive code into the design canvas — creatoroff · 2026-10-23
- Coding agents like Codex can already handle much of real-to-sim for robot manipulation — zhenjun_zhao · 2026-10-23
- Debate: open standards should keep agent runtimes portable, Kubernetes-style — matt_slotnick · 2026-10-23
- Lovart plugin goes live in Codex: generate brand visuals and ads where you code — OpenAIDevs · 2026-10-23
- Agent Moravec's law: billions of tokens to get Claude to symlink config files — yaroslavvb · 2026-10-23