Topic
The reliability, state and orchestration layer beneath long-running AI agents that call models, use tools and wait on humans or external systems.
OpenAI launched Dots at DevDay 2026, personal agents that run on cloud computers OpenAI maintains and keep context across ChatGPT, Slack and Teams. For unattended agent work, teams now choose between OpenAI's machines and hardware they can switch off themselves.
Reality
Scry's MCP search API prices an over-quota query instead of refusing it. That moves congestion control into the agent's retry loop, where a backoff timer cannot see the quote and the budget arbiter is the orchestrator's problem.
Reality
A dev.to post works through what the stateless-service playbook does to an agent request. The common probe cadence restarts a pod after roughly 30 seconds of silence on /health, and one slow tool call takes longer than that.
Reality
Ao Qu and collaborators open-sourced Reef on September 15th, an OpenAI-format inference server that stamps each response with a record ID, matches later feedback to that ID, and publishes a retrained artifact only if its evaluation stage accepts it.
Reality
Durable state and sandboxed execution now arrive as platform primitives, so a dev.to guide argues the remaining failure mode is undocumented organizational context. It leans on one arXiv paper and one example.
Reality
Alexandr Wang told a Y Combinator audience that a swarm of Meta agents scheduled by cron beat 100 engineers on specific tasks, and he credited the evaluation system for the result. Meta paid $14.3 billion for Scale AI.
Reality
Five named priorities now decide which specification proposals get read first, which makes the August roadmap a resource-allocation call for anyone building unattended agents.
Reality
A reported $500 million round at a pre-money floor of at least $12 billion reprices a two-decade systems bet. Teams already running workflows on it should treat this as a procurement event.
Reality