OpenAI just wrapped its annual developer conference, and for once the most interesting announcement wasn’t a new model. It was infrastructure. AI that keeps working after you close the tab.
The central release is GPT-6.1 Sol, an update to GPT-6 Sol built for coding, computer use, and professional tasks. It performs close to GPT-6 Astra, OpenAI’s most capable model, and costs one-fifth as much. Cached input tokens run at $0.10 per million. Developers testing early access were blunt about their reactions, with several describing something they rarely get from a model release: near-Astra quality output at pricing that makes production deployment at scale actually viable.
The price shift matters. Products that were too expensive to build six months ago are now economically reasonable.
The bigger news is computer use arriving in the Agents API. An AI agent can now operate software through a graphical interface, reading screens, clicking buttons, and navigating applications. This is not a demo capability. It is production infrastructure for building agents that complete digital tasks the same way a human operator would, except they do not clock out.
Dots takes this idea to its logical conclusion. These are persistent agents that continue working on tasks in the background without requiring you to start a new conversation each time. Assign Dots a job, and it runs. ChatGPT Space extends this to teams, creating a shared workspace where humans and agents operate around common context and history. For a business owner, that could mean an AI that manages customer intake, monitors a pipeline, or triages requests around the clock with nobody actively managing it.
The Decisions API, currently in limited preview, solves a specific friction point in AI workflow design. Routing requests, classifying inputs, and selecting the next action in an automated pipeline have always required custom code or brittle prompt logic. That friction compounds fast. A dedicated API using a fast, lightweight model called Luna makes structured decision-making something you configure rather than something you build from scratch every time.
OpenAI also expanded ChatGPT’s plugin system to support sidebar panels, interactive conversation interfaces, and custom file viewers built directly inside ChatGPT. Plugins can now respond to events from connected applications using the proposed MCP Events spec, which opens the door to automations that fire on their own when something changes in an external system.
The pattern across all of it is deliberate. OpenAI is building infrastructure for AI that runs operations, not just answers questions. The question for businesses is not whether AI can help anymore. It is which jobs should never come back to a human once handed off.
Want to explore how AI agents could benefit your business? Let’s talk.