Live from OpenAI DevDay: OpenAI turns ChatGPT into an agent platform

The headline: ChatGPT wants to become the place where agents work

OpenAI’s DevDay 2026 made one strategic message clear: the next phase of ChatGPT is not only about answering prompts, but about hosting persistent agents that can take on work, coordinate with software and stay active across a user’s day. Axios described the event as a slate of more than 20 announcements aimed at moving AI “beyond the chatbot” and into software that can act independently on users’ behalf [1]. That framing matters because it changes ChatGPT’s role from a conversational interface into something closer to an operating layer for AI-assisted work.

The symbolic center of the keynote was Dots, OpenAI’s always-on agent product. OpenAI’s own deployment safety documentation says Dots are built on existing model and agent capabilities, powered by GPT-6 Astra, and designed to handle recurring work with their own cloud computer and browser [2]. That is a sharp product shift: instead of asking a model a question, the user can delegate an ongoing responsibility to an agent that has memory, tools and a place to run.

The scale behind that ambition is also part of the story. Reporting from The Next Web says OpenAI told press that ChatGPT now has 1.2 billion weekly users, while more than 35 million people use ChatGPT Work and Codex weekly [3]. Those numbers give developers a reason to pay attention: if ChatGPT becomes a distribution surface, a workflow layer and an agent runtime at the same time, building “inside ChatGPT” starts to look less like a feature bet and more like a platform bet.

Dots: the personal agent as a product category

Dots are OpenAI’s answer to a problem every major AI company is now trying to solve: how to move from reactive chat to proactive execution without overwhelming users. The AP reported that Sam Altman introduced a “remarkably capable, always-on” agent at DevDay, designed to complete ongoing tasks proactively on behalf of users [4]. OpenAI’s safety documentation describes each dot as having its own cloud computer and browser, using connected tools, handling recurring work and following up across channels such as ChatGPT, text message, email and Slack [2].

That product design is powerful, but it also creates a new trust burden. A chatbot that makes a mistake in a single answer can usually be corrected in the next turn. An agent that works in the background can make mistakes across files, calendars, repositories, messages or customer workflows. OpenAI appears to understand that distinction: the same safety documentation frames Dots as a public launch that required additional safeguards and review [2]. In other words, the product OpenAI wants to sell is not just intelligence; it is bounded delegation.

The early availability also shows that OpenAI is not treating Dots as a universal free feature. Axios reported that Dots are available to Pro, Business and Enterprise users, with one dot per user to start [4]. That tiering suggests OpenAI is beginning with higher-value users whose use cases justify the cost of running persistent agents in the cloud.

ChatGPT Space, Codex and the team workflow push

DevDay was not only about personal agents. OpenAI also pushed ChatGPT deeper into team collaboration. The Next Web reported that ChatGPT Space replaces Library as a shared home for team files and documents, with pages that teammates, ChatGPT and Dots can edit together [3]. It also reported that OpenAI is bringing collaborative slides, Team Tasks, Slack and Microsoft Teams mentions, and meeting-note features into the same work environment [3].

This is important because agentic AI needs shared context. A personal assistant can help an individual, but a work agent becomes more valuable when it can see documents, team decisions, project status and code. By connecting Dots, Space and Codex, OpenAI is trying to make ChatGPT less of a destination app and more of a daily workplace.

Codex also remains central to the developer strategy. The Next Web reported that Codex can now run coding tasks in the cloud, continue while a developer’s computer is asleep and support code review workflows connected to GitHub or GitLab [3]. That gives developers a practical reason to test OpenAI’s agent stack: coding is one of the few places where agentic work can be evaluated quickly, through builds, tests and diffs.

The pricing signal: faster agents cost real money

The most commercially revealing part of DevDay may have been pricing. The Next Web reported that OpenAI launched a $500-per-month Pro 500 plan with access to its fastest speed tier, while changing the usage allowance of the $200 Pro plan from October 30 [3]. That is a reminder that always-on agents are not cheap. They require inference, storage, tool access, monitoring and cloud computers that keep running after the user walks away.

The pricing move also gives shape to OpenAI’s broader business model. Axios reported that OpenAI’s annual recurring revenue is nearing $70 billion, with enterprise sales more than doubling since July, according to sources familiar with the financials [7]. Whether that figure is ultimately confirmed in audited filings or not, the direction is clear: OpenAI is trying to convert massive consumer reach into paid work, developer and enterprise usage.

For users, this creates a practical question: will agents save enough time to justify premium pricing? For businesses, the question is different: can these agents be controlled, audited and integrated safely enough to become part of normal operations? DevDay’s answer was to offer more infrastructure, more admin controls and more workspace features, not just a smarter model.

AMD, World Labs and the infrastructure race behind the keynote

The same 72-hour news cycle around DevDay also showed why OpenAI’s agent strategy depends on more than software. AMD announced that it had entered a definitive agreement to acquire World Labs, the spatial-intelligence startup led by Fei-Fei Li, in an all-stock transaction valued at about $8.2 billion [5]. AMD said World Labs develops models that generate, reconstruct and simulate interactive 3D environments from text, image and video inputs, and that the deal is expected to close by the end of 2026 subject to approvals [5].

That acquisition is not an OpenAI announcement, but it belongs in the same market story. Agents that act in software today may need richer world models tomorrow: simulation, robotics, physical AI and 3D reasoning all require different compute patterns. AMD is buying model expertise to inform future AI infrastructure, while OpenAI is launching the agent layer that will consume that infrastructure.

Oura’s delayed IPO and the market’s mixed mood

The broader AI-adjacent market is not uniformly euphoric. AP reported that smart-ring maker Oura put off its IPO because of market uncertainty, with concerns around AI spending, interest rates and bond yields weighing on the decision [6]. At the midpoint of its expected range, the IPO would have valued Oura at $13.5 billion [6].

That delay is a useful counterpoint to DevDay. OpenAI can present agents as inevitable, AMD can spend billions on world-model talent, and investors can still ask whether the market is overheating. The lesson is not that AI demand is weak. It is that the next wave must prove usefulness, not just novelty.

The bottom line

OpenAI DevDay 2026 was a platform announcement disguised as a product blitz. Dots gives ChatGPT a persistent agent layer. Space gives teams a shared context layer. Codex gives developers a measurable agent workflow. Pricing gives OpenAI a path to monetize heavy usage. And the surrounding AMD and Oura news shows the market testing both the infrastructure upside and the valuation risk of the AI cycle.

The future v2.42 is not arriving as one magical model. It is arriving as agents, workspaces, pricing tiers, safety reviews, cloud computers and distribution channels. OpenAI’s bet is that users do not want more AI content; they want less unfinished work.

Sources from the last 72 hours

1. [1]The 5 biggest announcements from OpenAI's blockbuster AI conferenceSep 29, 2026, 8:22 PM
2. [2]GPT-6 Astra System Card - OpenAI Deployment Safety HubSep 29, 2026, 2:00 AM
3. [3]OpenAI halves Pro 200 usage and launches a $500 ChatGPT plan at DevDaySep 29, 2026, 7:58 PM
4. [4]Altman unveils 'always-on' AI agent after OpenAI shelves model over safety concernsSep 29, 2026, 8:51 PM
5. [5]AMD to Acquire World Labs to Advance the Future of AI ComputeSep 28, 2026, 2:00 AM
6. [6]Smart ring maker Oura puts off initial public offering due to market 'uncertainty'Sep 29, 2026, 3:57 PM
7. [7]Scoop: OpenAI's annual recurring revenue nears $70BSep 29, 2026, 4:15 PM

AI-generated article based on recent web research, then preserved as a dated editorial snapshot.