OpenAI DevDay 2026 arrived with more than 20 major announcements across ChatGPT, Codex, models and developer tools, but the most important story was not a single model launch. GPT-6.1 Sol is certainly part of the announcement, yet the larger direction is much more ambitious: OpenAI is trying to turn AI from something people prompt occasionally into a persistent working layer where agents can operate in the background, developers can build native experiences inside ChatGPT, and teams can work alongside AI in shared environments. OpenAI says ChatGPT now reaches 1.2 billion weekly users, giving that developer ecosystem an unusually large distribution surface.
That makes DevDay 2026 feel less like a conventional product refresh and more like an architecture shift. The company is connecting several previously separate pieces—models, agents, Codex, plugins, APIs, cloud execution, shared workspaces and third-party applications—into a more unified system. The result is a vision in which the model is no longer the entire product. The model becomes one component inside a larger environment capable of reasoning, using tools, taking action, maintaining context and handing work back to people.
The Big Shift: From One-Prompt AI to AI That Keeps Working
The clearest expression of this idea is Dots, OpenAI's new always-on agents. OpenAI describes Dots as agents that can work toward a user's goals continuously, learn preferences from feedback and operate through their own cloud computers. They can connect to more than 4,000 apps through OpenAI's plugin ecosystem and can be reached through ChatGPT, Slack and Microsoft Teams.
The difference from a normal chatbot is not simply that Dots can generate better answers. Their value comes from continuity. OpenAI's examples include following a product launch, updating research when new evidence arrives, maintaining an evolving sales proposal and preparing content when a new transcript appears. In each case, the user establishes the direction and the agent continues working as information changes rather than waiting for a fresh prompt every few minutes.
That changes the basic interaction model. A traditional assistant is largely request → response. An agentic system is closer to goal → ongoing work → review → next action. OpenAI is clearly betting that this second model will become more useful for professional workflows, particularly when the work involves many applications, repeated decisions or information that changes over time. Reuters also described Dots as part of OpenAI's growing push toward autonomous AI for both consumers and enterprise users.
Codex Is Moving From a Coding Assistant to a Development Environment
The second major piece of the strategy is Codex. OpenAI introduced Codex in the cloud as a way to run development work from a computer, remotely from a phone or directly in the cloud. The company is also introducing reusable development environments with shared settings and permissions, allowing tasks to start with a more consistent setup rather than recreating the environment every time.
For developers, this is more significant than simply giving an AI better coding abilities. The environment around the model matters because real software work involves repositories, tools, credentials, testing, review, security checks and long-running tasks. OpenAI is therefore building more of that infrastructure around Codex itself. The updated CLI adds voice control and an /agents view for delegating and tracking tasks, while the new Code Review experience can summarize changes, inspect diffs and run automatic reviews in the cloud.
OpenAI is extending that idea into security as well. Codex Security Cloud can scan repositories, investigate findings, remove duplicates and prepare fixes, with scans capable of continuing in the cloud even when a developer's laptop is closed. That makes Codex look less like a code-generation feature and increasingly like an AI-operated development workflow.
TechCrunch noted a similar shift in its coverage of the announcement: Codex's cloud environments are becoming more persistent and configurable rather than simply acting as isolated remote execution sandboxes.
The Agents API Is the Infrastructure Behind the Idea
OpenAI's broader developer strategy becomes clearer when Codex is connected to the Agents API. The API, introduced publicly in September, is designed to provide developers with the same kind of harness OpenAI uses to run long-running agents: context management, tool use, subagents and reliable cloud execution. At DevDay, OpenAI extended that system with computer use, allowing agents to interact with software interfaces rather than being limited to text or API calls.
This is a crucial architectural change. Earlier AI applications often looked like a thin layer around a model: send a prompt, receive a response and perhaps call an API. The newer architecture needs an execution layer around the model. The agent needs to decide which tool to use, maintain state, delegate work, access files, interact with software and sometimes operate for hours rather than seconds.
OpenAI is also bringing these capabilities into other infrastructure. Through Bedrock Managed Agents, powered by OpenAI, developers can build OpenAI-powered agents that operate inside AWS and integrate with AWS resources. The result is a more portable agent stack in which OpenAI's intelligence can sit inside different cloud environments rather than forcing every workflow into one interface.
ChatGPT Is Becoming a Platform, Not Just a Product
Perhaps the most strategically important announcement for developers is Plugin Extensions. OpenAI is opening the platform it uses to build ChatGPT features so developers can create experiences that feel native inside ChatGPT. A plugin can have a dedicated sidebar location, interactive panels beside a conversation and its own file viewers. OpenAI's developer documentation now describes these extensions as a way to make plugins feel like first-class ChatGPT features.
This matters because distribution has always been one of the hardest problems for developers. Building a useful tool is only half the challenge; getting users to discover it, install it and understand how to use it is another. OpenAI is effectively turning ChatGPT into a distribution layer where outside developers can bring software closer to the user's conversation and workflow.
OpenAI is also making plugins more event-driven. With support for the proposed MCP Events specification, a plugin can initiate an automation when something happens in a connected application. A new project task, for example, could trigger ChatGPT to read related documents and draft a plan without waiting for the user to issue another prompt.
That is an important evolution from “AI that responds” to AI that reacts to events.
ChatGPT Space Changes Where the Work Happens
Then comes ChatGPT Space, which ties the agent and application strategy together. OpenAI describes Space as a shared environment where teammates, ChatGPT and Dots can work from common knowledge and project context. It is designed to keep files and pages together, organize shared material and make it easier for teams to continue work instead of restarting conversations from scratch.
The companion feature, Pages, pushes the concept beyond a chat window. Pages are editable documents designed for human-and-agent collaboration, allowing users to write, research, generate charts, create images and visualize information inside the same working environment. OpenAI is also planning collaborative slides that multiple teammates and agents can edit together.
This is where the announcements begin to connect. Dots provide persistent agents. Codex provides a development execution layer. Plugins connect external software. Space provides shared context. Pages provide an editable working surface. The model sits underneath all of them, but it is no longer the only thing developers are building around.
GPT-6.1 Sol Matters — But It Supports the Larger Architecture
DevDay would not be complete without a model announcement, and OpenAI introduced GPT-6.1 Sol, an upgrade to GPT-6 Sol aimed particularly at agentic coding, computer use and professional workflows. OpenAI says it approaches GPT-6 Astra's performance on several evaluations while costing one-fifth of Astra's standard input and output token prices. The company also announced an Ultrafast tier capable of up to 8× faster token generation in Codex and up to 6× faster generation in the API.
The significance is less about model naming and more about economics. Agents can be expensive because they may make many model calls, use tools repeatedly and maintain context across long workflows. Lower inference costs therefore directly affect whether developers can run more ambitious agentic systems at scale. A cheaper model that is sufficiently capable for coding, computer use and professional workflows can change the economics of an entire application architecture.
OpenAI's own positioning reinforces that point: GPT-6.1 Sol is being offered as a way to execute capable workflows within the same budget rather than simply as a larger benchmark number.
Knowledge Corner: What Actually Changed at DevDay?
Agentic AI means AI systems that can pursue a goal through multiple steps, use tools and take actions rather than only generating a response.
Agent harness refers to the surrounding infrastructure that helps an AI agent manage context, tools, execution, subagents and long-running tasks. OpenAI describes the Agents API as bringing the Codex harness and infrastructure to developers.
Computer use allows an agent to interact with software through interfaces rather than relying only on structured APIs. At DevDay, OpenAI added computer use to its Agents API.
MCP Events is a proposed specification that allows connected systems to trigger actions when events occur, enabling more event-driven agent workflows. OpenAI is adding support for it in plugin automations.
Why Developers Should Pay Attention
For developers, the biggest lesson from DevDay 2026 is that the competitive advantage may no longer come from simply choosing the newest model. The application architecture around the model is becoming equally important. Developers now have access to hosted agent infrastructure, computer use, multi-agent workflows, cloud execution, plugin interfaces and shared AI workspaces.
That changes what a strong AI application looks like. A simple chatbot might answer a customer question, but a more advanced system could monitor incoming requests, classify them with the Decisions API, call the right tools, update a shared workspace, ask for approval when a sensitive action is required and continue working after the original conversation has ended. The model is still important, but orchestration, permissions, tools, context and execution become first-class engineering concerns.
For students and early-career developers, this creates an unusually practical opportunity. Instead of building another “ask my PDF” application, a stronger portfolio project could combine an agent, MCP tools, a persistent task workflow, human approval and Codex-style code execution. The goal should be to demonstrate that you understand how an AI system actually operates in the real world—not just how to send a prompt to an API.
The Bigger Story: OpenAI Wants ChatGPT to Become the Place Where AI Work Happens
Taken individually, Dots, Codex Cloud, Plugin Extensions, Space, Pages and the Agents API look like separate announcements. Together, they reveal a much clearer strategy.
OpenAI is trying to make ChatGPT the surface where users, developers, applications and agents meet. Developers can build experiences inside ChatGPT. Agents can use those experiences. Teams can collaborate with agents inside Spaces. Codex can execute development work in the cloud. Plugins can connect external applications. Events can trigger automated workflows. And lower-cost models can make those workflows more economically viable. OpenAI itself says DevDay is about opening ChatGPT as a shared surface for humans and agents and allowing developers to launch native experiences to its 1.2 billion weekly users.
That is why calling DevDay 2026 simply “the GPT-6.1 Sol launch” misses the larger picture. The model is one part of the announcement, but the more consequential development is the attempt to build an AI operating environment around models—one where software does not merely answer questions but can continue work, use tools, collaborate with people and respond to events.
The next stage of AI may therefore be defined less by which model answers the prompt and more by what the entire system can do after the prompt is finished.
Official Sources & Further Reading
OpenAI — DevDay 2026 Recap
OpenAI — Introducing Dots
OpenAI — GPT-6.1 Sol
OpenAI — Introducing the Agents API
OpenAI Developers — Plugins
