OpenAI, Anthropic, Cursor All-In on Cloud Agents
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- OpenAI, Anthropic, and Cursor are all-in on cloud agents, with engineers at all three labs telling the author that demand for hosted, long-running agents is set to increase massively
- Anthropic's "Claude Managed Agents" — a hosted service built over six months by Katelyn Lesse's Claude Platform team — runs long-running agents across multiple cloud providers
- OpenAI acquired Ona (formerly Gitpod) to power cloud development environments for Codex agents and is actively hiring engineers for its Cloud Agents team
- Cursor cofounder and CPO Sualeh Asif detailed engineering challenges unique to cloud agents, including a "confessions" mechanism where models periodically report environment issues to the infra team; the company launched a cloud-agent-powered iOS app on June 29
- OpenAI's internal data shows more than 95% of non-engineers use Codex rather than ChatGPT, per first Codex engineer Andrew Ambrosino — a sign coding harnesses may spread well beyond developers
- The cloud-agent shift is enabled by coding models reaching "good enough" autonomous quality (Opus 4.5 / GPT-5.4), matured MCP and skills infrastructure, context windows of up to 1 million tokens, and expanded GPU capacity across cloud providers
Why it matters: The locus of agent work is moving off engineers' laptops and into managed cloud infrastructure, which changes what engineering teams build — Cursor's Asif says more work is now about making agent environments more efficient than writing code directly. All three labs are simultaneously redirecting hiring and product roadmaps around the same thesis, with OpenAI's Ona acquisition, Anthropic's six-month Managed Agents build, and Cursor's iOS launch as concrete proof points.
.jpg)