OpenAI used DevDay 2026 to expand its developer tools for agents, including computer use in the Agents API and cloud-based Codex environments. The practical shift is towards agents that can operate software and run coding tasks remotely, not just answer a prompt and wait politely for the next one.
OpenAI Watch analysis
What happened
OpenAI also announced GPT-6.1 Sol, a Decisions API in limited preview, new ChatGPT plugin capabilities and Dots, persistent agents for ongoing tasks. InfoQ’s 2 October recap says the Agents API can use graphical interfaces and incorporates tool search, tool calling, multi-agent capabilities and context compaction. Availability is described as limited to selected plans in some cases.
For developers, cloud Codex can run tasks away from a local computer, while the Decisions API uses a smaller model to select from predefined answers based on text or image context. InfoQ says GPT-6.1 Sol is aimed at coding, computer use and professional work; OpenAI says it approaches GPT-6 Astra on several evaluations at one-fifth of Astra’s standard input and output token prices. That is the company’s comparison, not a like-for-like performance result established in the recap.
Why it matters
Taken together, the announcements point towards software that can act across tools, interfaces and longer-running tasks. Computer use and remote coding environments could make agent workflows more useful to developers; the trade-off is that a tool capable of operating software needs clearer boundaries than a chatbot producing a tidy paragraph.
The lower-priced model and constrained Decisions API also offer different ways to put AI into products: one targets capable general work at a lower stated price, the other selects from a predefined set of answers. These are distinct approaches, not proof that every workflow needs an agent with its own computer.
Our read
The most consequential change is the move from individual model calls towards infrastructure for agents that can use tools and keep working. Developers should look first at what the Agents API and cloud Codex actually let them do, then test the permission and reliability questions in their own workflows. The breadth is notable; the recap gives less detail on implementation limits, so the launch list is a starting point, not a deployment plan.
What to watch
- Which plans can use computer use, cloud Codex and the new plugin features.
- What safeguards and permission controls apply when agents operate software.
- How GPT-6.1 Sol performs on developers’ workloads at its stated prices.
- When the Decisions API moves beyond limited preview.
Discussion spark: Should developers let agents operate software through graphical interfaces, or keep them on narrower, explicitly defined tools?
Sources and evidence
- OpenAI DevDay 2026 Recap for Developers – infoq.com (2 October 2026, 10:39 UTC)
OpenAI Watch is independently operated by WittyWires. It is not affiliated with, endorsed by, or operated by OpenAI.