In brief: OpenAI used DevDay 2026 to put persistent AI agents at the center of its platform, introduce GPT-6.1 Sol and expand Codex, ChatGPT and its developer infrastructure. The key takeaways are a public beta of the Agents API, shared ChatGPT workspaces, approved app plugins, cloud-based coding environments and a new $500 Pro plan for intensive users. Together, these product launches point toward a future in which people delegate more complex work to AI while retaining oversight at important steps.
OpenAI’s annual developer event took place in San Francisco on September 29, against a backdrop of growing concern about security incidents involving advanced models and AI agents. The company called this its biggest DevDay yet, presenting around 20 major AI announcements across ChatGPT, Codex, GPT models and developer tools. The central theme was not simply more capable software, but new ways for people and AI systems to collaborate over time.
That ambition comes with practical questions: how should agents access tools, who approves consequential actions, and how can teams keep work organized? OpenAI’s announcements offer part of an answer through persistent agents, permissioned workspaces and cloud environments designed for delegated tasks. For a small software studio, for example, that could mean assigning an agent a coding task, reviewing its progress and approving a sensitive action rather than supervising every step. The broader message is that AI is moving from a prompt-and-response tool toward a working partner—making the design of safeguards as important as the capabilities themselves.
OpenAI DevDay 2026 puts persistent AI agents in focus
One of the event’s defining announcements was Dots, always-on agents based on GPT-6 Astra. Unlike an assistant that waits for a new prompt, these agents are intended to remain active and take responsibility for ongoing work. That could help a team monitor recurring tasks or keep a project moving between check-ins, while making clear rules about access and oversight essential.
OpenAI also opened the Agents API in public beta. It includes computer use, compaction to manage extended tasks, and multi-agent orchestration, allowing developers to build systems that use tools and carry out sequences of actions; validation remains necessary at important stages. For a beginner exploring AI agents, the practical distinction is simple: an agent can do more than answer a question, but its permissions should match the consequences of the work.
The shift is significant, but it does not mean handing over every decision. A useful way to think about these systems is as capable assistants with defined responsibilities: the more consequential the task, the clearer the checkpoints should be. That balance between autonomy and human review is the foundation for the announcements that follow.
GPT-6.1 Sol and Codex Cloud reshape software development
OpenAI introduced GPT-6.1 Sol, an evolution of GPT-6 Sol aimed especially at agentic coding, computer use and professional tasks. The company also highlighted GPT-6 Astra Ultrafast as a high-speed option for users who need faster generation. For developers, the focus is on models that can work through a task and interact with software, not just produce text or code on request.
Codex is evolving in the same direction. With Codex Cloud, developers can delegate programming work to isolated environments hosted in the cloud, keeping an agent’s task separate from their local setup. Imagine a developer asking Codex to investigate a bug in a contained workspace; they can inspect the result before deciding whether to bring it into the main project. These developer tools could make delegation more practical, provided teams continue to review and test the output.
For people new to AI-assisted coding, the important change is the workflow: describe a goal, let the system work in a bounded environment, then review what it produces. The cloud sandbox is not a substitute for testing, but it gives developers a clearer place to delegate. In short, Codex is being positioned less as a code autocomplete feature and more as a managed development agent.
ChatGPT Space and plugins bring AI collaboration into the workplace
ChatGPT Space is a shared environment where colleagues and agents can work together. Teams can assign recurring tasks and set permissions and responsible owners, giving people a way to coordinate delegated work rather than leaving it scattered across individual chats. For a small marketing team, a Space could bring a recurring research task and its human reviewer into the same shared workflow.
OpenAI also emphasized plugins that let businesses connect approved external applications directly to ChatGPT. Canva, Figma and Adobe were among the examples named. These integrations could help users move between planning, design and content tasks without switching contexts as often, while the approval requirement gives organizations a way to control which services are available.
OpenAI said developers could launch new native experiences to its audience of 1.2 billion users. That scale may make the platform attractive to builders, but it also makes thoughtful permissions and clear user expectations especially important. The strongest workplace use cases will pair convenience with visible control over what an agent or plugin can access.
OpenAI’s $500 Pro plan and the wider DevDay outlook
For users with especially intensive workloads, OpenAI announced a $500 Pro subscription that includes access to GPT-6 Astra Ultrafast and removes the five-hour limit described for that access. The plan is aimed at people who rely on the service heavily, rather than casual users who may not need its higher-capacity offering. OpenAI also announced a software marketplace for businesses, extending the idea of bringing tools and services into its ecosystem.
These launches arrive as the industry faces heightened scrutiny over AI-agent security. OpenAI founder Sam Altman had confirmed earlier in September that the company’s long-anticipated IPO would not happen this year, while the business continues to invest in products and infrastructure. DevDay therefore offered a picture of both ambition and pressure: more capable systems are being introduced at the same time that companies must show how those systems can be governed responsibly.
For developers deciding what to explore first, a sensible starting point is to match the tool to a contained, repeatable task: test an agent in a limited environment, define who approves important actions, and measure whether it genuinely saves time. Across its API updates, GPT models and collaboration features, the clearest signal from DevDay is that AI is becoming an active part of software development and everyday work—but human judgment remains part of the design.
