OpenAI's 'dots' always-on agents: what DevDay means for anyone commissioning AI software
OpenAI's DevDay introduced dots, always-on AI agents, plus an Agents API with computer use and a cheaper GPT-6.1 Sol. Founders searching for AI agents, AI app development and Codex news should read this as a build-cost shift, not a product launch.
7 October 2026
OpenAI’s DevDay on 29 September 2026 produced more than 20 announcements, and searches for “OpenAI dots”, “always-on AI agent” and “Agents API computer use” have been climbing since. Three of those announcements matter if you are planning AI software, rather than just using it.
What was announced
- Dots: persistent agents that keep working on ongoing tasks rather than answering one prompt and stopping. Reporting so far gives little detail on pricing or full availability, so treat the specifics as unconfirmed.
- Agents API with computer use: applications can now have an agent operate other software through its on-screen interface, with multi-agent support, tool search and context compaction handled by OpenAI’s infrastructure.
- GPT-6.1 Sol: priced at $2 per million input tokens and $10 per million output tokens, and described by OpenAI as close to its top GPT-6 Astra model on agentic coding and computer use at roughly one-fifth of the price.
- Codex in the cloud: coding tasks can run remotely and be started from another device, with a new interface for delegating work across several agents.
The interpretation
The headline is not a single product. It is that the plumbing for agents, the part that used to take a development team months to build, is now rented from the model vendor. Running an agent for hours, letting it use a legacy system with no API, and keeping its memory tidy are becoming checkbox features.
That lowers the cost of a first working version. It does not lower the cost of getting it right: deciding what an agent is allowed to do unattended, how its mistakes are caught, who approves risky actions, and what happens when a vendor changes a model or a price. An always-on agent with the wrong permissions is a bigger problem than a chatbot with the wrong answer.
So what
If you have been quoted a large figure for an agent-based product, ask which parts are now vendor-provided and re-price the rest. If you have not started, pick one narrow, reversible workflow and prove it before committing to a platform. BuildApps designs AI products with approval steps, logging and model switching built in, as part of AI-assisted development. To scope an agent project, start a project.