OpenAI just gave one of its clearest looks yet at what it thinks comes after the chatbot.
At DevDay 2026, held September 29, OpenAI announced more than 20 updates spanning ChatGPT, Codex, its API, plugins, enterprise tools and entirely new ways for people to work alongside AI.
The biggest headline was Dots, OpenAI's new class of always-on AI agents. But taken together, the announcements point to something much larger.
OpenAI is increasingly turning ChatGPT into a platform where people, AI agents, apps and company data can work together continuously—not just a place where you type a prompt and wait for an answer.
OpenAI also revealed that ChatGPT has now reached 1.2 billion weekly users, making the platform increasingly significant not only as an AI assistant, but potentially as a distribution channel for developers building software around it.
Here are the announcements that matter most.
The centerpiece of DevDay was Dots.
OpenAI describes Dots as always-on agents designed to take responsibility for ongoing work rather than waiting for individual prompts.
A Dot can learn what matters to you, use connected tools and continue working on tasks over time. OpenAI is initially making Dots available on Pro and Business Premium, while Enterprise, Education and Healthcare customers can enable a beta through administrators.
This represents a significant shift from the traditional chatbot interaction.
Instead of:
Ask → Answer → Conversation ends
the model becomes closer to:
Assign responsibility → Monitor → Act → Report back
Reuters described the launch as part of OpenAI's broader push into autonomous agents and enterprise AI, while noting that Dots can operate across applications and communicate through services including Slack and Microsoft Teams.
That distinction matters.
The future OpenAI demonstrated at DevDay is not simply about making ChatGPT answer questions better. It is about giving AI responsibility for pieces of work.
OpenAI also launched GPT-6.1 Sol, an upgrade focused heavily on agentic coding, computer use and professional work.
According to OpenAI, GPT-6.1 Sol delivers performance approaching its more powerful GPT-6 Astra model while costing just one-fifth of Astra's standard input and output token prices.
That pricing may prove especially important for developers.
The economics of agents are different from the economics of chatbots. An agent might make dozens of model calls, search documents, use tools, operate software and continue working for extended periods.
As AI applications become more autonomous, inference cost quickly becomes part of the product architecture.
A model that approaches frontier performance at substantially lower cost therefore gives developers more room to build sophisticated workflows without making every agent interaction prohibitively expensive.
TechCrunch reports that OpenAI is positioning GPT-6.1 Sol particularly around programming, debugging, document understanding and complex multi-step workflows.
Another major DevDay announcement was the expansion of the Agents API.
The API now supports computer use, allowing developers to create agents capable of interacting directly with software.
OpenAI is also bringing capabilities associated with Codex—including multi-agent orchestration, tool search, tool calling and context compaction—into applications built with the Agents API.
Perhaps more importantly, OpenAI says it will operate the underlying infrastructure itself.
That puts OpenAI in an interesting position.
It doesn't just want to supply the intelligence inside agents.
It increasingly wants to supply the runtime that agents operate on.
For developers, that could mean less time assembling separate browser automation systems, orchestration frameworks, model infrastructure and execution environments.
OpenAI also introduced a Decisions API.
Rather than asking a general-purpose model to produce an open-ended response, developers provide a set of predefined choices and context in text or images.
The system then chooses among those options.
Potential applications include:
The Decisions API launched in limited preview, with broader availability planned shortly after DevDay.
It may sound less exciting than an autonomous agent, but it addresses an important problem: not every decision inside an AI workflow requires an expensive frontier model generating paragraphs of text.
Sometimes the system simply needs to decide what happens next.
One of the most strategically important announcements received less attention than Dots.
OpenAI is expanding what developers can build through ChatGPT plugins.
New Plugin Extensions can have their own presence in the ChatGPT sidebar, interactive panels and custom viewers for supported file types. OpenAI is effectively opening more of the same platform it uses to build native ChatGPT functionality to outside developers.
OpenAI is also improving plugin creation, submission and discovery.
A new Plugin Creator is designed to make plugins easier to build, while recommendations can surface relevant plugins both in the plugin directory and directly within ChatGPT conversations.
Users remain in control of which plugins they connect and what permissions those plugins receive.
This could become a significant distribution shift for AI software companies.
For years, software discovery revolved around search engines, app stores, social media and marketplaces.
Now another possibility is emerging:
What happens when the AI assistant itself decides that a particular tool is relevant to what the user is trying to accomplish?
For AI startups, being discoverable to agents may eventually matter alongside being discoverable to humans.
Plugins are also becoming event-driven.
OpenAI announced support for the proposed MCP Events specification, allowing connected apps to trigger automations when something changes.
For example, OpenAI says a user could ask ChatGPT to monitor a project board. When a new task appears, ChatGPT could read the associated documents and prepare a plan—even if the user isn't currently sitting inside ChatGPT.
That creates an important architectural shift.
Traditional AI:
You prompt the AI.
Event-driven AI:
Something happens → AI notices → AI starts working.
Combine that with persistent agents like Dots and the direction becomes clear.
AI is becoming less reactive.
OpenAI also introduced ChatGPT Space, designed as a collaborative environment for teams, ChatGPT and Dots.
A Space can contain shared knowledge that both humans and AI agents build upon, while ChatGPT can help organize that information based on instructions from the team.
It will be available to Pro, Business and Enterprise customers across desktop and web, with some functionality also available on mobile.
That moves ChatGPT closer to being a workplace rather than simply an assistant.
OpenAI's new Pages feature continues that theme.
Pages are collaborative documents where people and AI can work together on writing, research, charts, images and other content.
Users can create a Page from a conversation and invite teammates to contribute or review it.
The concept competes less with a conventional chatbot and more with productivity platforms such as collaborative document editors and knowledge workspaces.
OpenAI also announced collaborative presentation creation.
Users will be able to create slides from conversations or existing templates while multiple teammates and AI agents edit the presentation simultaneously.
The resulting presentations can be shown inside ChatGPT or exported to PowerPoint and Google Slides while preserving formatting.
This matters because OpenAI is moving beyond generating an artifact for the user.
It increasingly wants ChatGPT to be the environment where the artifact is created, edited, reviewed and shared.
For Business and Enterprise customers, ChatGPT teams can now share Pages, slides, plugins and spreadsheets.
Teams can also delegate recurring work.
For example, a team might ask ChatGPT to prepare a weekly project update. ChatGPT can use connected tools to gather the information and run the task according to a schedule or in response to an event such as an incoming email or Slack message.
This begins to blur the line between generative AI and traditional business-process automation.
OpenAI is also allowing Business and Enterprise customers to call @ChatGPT directly from Slack and Microsoft Teams.
The assistant can work inside channels, threads and direct messages while accessing permitted tools and maintaining the context supplied by participants.
For enterprises, that reduces one of the biggest barriers to AI adoption: asking employees to move their existing work into yet another application.
Instead, ChatGPT increasingly comes to where the work is already happening.
OpenAI's new Meetings plugin takes notes, produces personalized summaries and action items, and stores them inside ChatGPT Space.
Those notes can then become inputs for additional work—such as updating project plans or drafting follow-up communications.
OpenAI says meeting audio is deleted once the notes are prepared and cannot subsequently be accessed or replayed.
Developers also received a substantial collection of Codex updates.
Reusable cloud development environments allow Codex tasks to run from a computer, phone or cloud environment with predefined settings and permissions.
The Codex CLI can now accept voice instructions and includes a new /agents view for monitoring multiple delegated tasks.
A new Code Review experience allows developers to inspect diffs and ask Codex questions about potential problems, while automatic reviews can run remotely.
And Codex Security Cloud can continuously scan GitHub repositories, investigate findings, remove duplicates and prepare fixes even while a developer's computer is offline.
Again, the pattern is persistence.
OpenAI wants its software agents to keep working when the human isn't.
OpenAI also announced Bedrock Managed Agents powered by OpenAI, developed with Amazon.
The service brings capabilities from OpenAI's Agents API into AWS so organizations can build OpenAI-powered agents that run inside AWS and integrate with AWS resources.
For large enterprises with strict infrastructure and governance requirements, that could make deploying OpenAI agents easier without moving core systems outside their existing AWS environments.
Greater agent autonomy also creates bigger privacy questions.
OpenAI introduced Private Intelligence, including Zero Data Retention with Private Safety Processing.
The company says this allows automated safety reviews without giving OpenAI personnel access to the underlying content.
A Private Inference preview planned for later this year will combine confidential computing with additional verifiable controls.
As agents gain access to corporate documents, email, code and internal applications, these privacy guarantees may become as important as model intelligence itself.
OpenAI also unveiled Ultrafast, a premium compute tier aimed at workloads where latency matters.
OpenAI says GPT-6 Astra Ultrafast can generate tokens up to eight times faster in Codex—reaching around 300 tokens per second—and up to six times faster through the API.
GPT-6 Astra Ultrafast is initially available through the API and to eligible Pro 500 and Enterprise customers, with GPT-6.1 Sol Ultrafast planned later.
OpenAI is effectively beginning to price not just intelligence and usage, but speed itself.
OpenAI introduced Pro 500, which offers its highest individual usage allowance at 25 times the ChatGPT Plus allowance and includes Ultrafast access.
The existing $200 Pro plan remains part of OpenAI's offering, while the new tier targets people whose AI usage increasingly resembles professional compute consumption rather than occasional chatbot use.
OpenAI also wants your ChatGPT subscription to travel with you.
Its new Sign in with ChatGPT system lets eligible users use parts of their ChatGPT plan allowance across participating third-party products.
OpenAI says 16 initial partners include Devin, Notion, Vercel, T3, OpenClaw and Dactyl.
This could become strategically important.
If developers can tap into a user's existing OpenAI subscription rather than requiring that person to separately pay for every unit of AI inference, it changes how some AI applications could be distributed and monetized.
The company also announced the OpenAI Marketplace for eligible enterprise customers.
Businesses will be able to apply part of their existing OpenAI spending commitments toward approved software from partner companies.
Initial partners include Figma, Adobe, Salesforce, ServiceNow, HubSpot, CrowdStrike and others.
That is another indication that OpenAI increasingly sees itself as an ecosystem platform rather than simply an AI model provider.
It would be easy to summarize DevDay as:
OpenAI launches Dots and GPT-6.1 Sol.
But that misses what happened.
The pieces increasingly fit together:
Models provide intelligence.
Agents provide autonomy.
Plugins provide access to software and data.
MCP Events tell the agents when something has happened.
Dots provide persistence.
Pages, Spaces and Slides give humans and agents places to collaborate.
Codex brings the same model to software engineering.
Slack and Teams bring ChatGPT into existing workplaces.
And OpenAI now has a potential distribution surface of 1.2 billion weekly ChatGPT users.
The company is no longer competing only to build the smartest chatbot.
It increasingly appears to be building an operating layer for AI work.
For developers and AI startups, that creates a new strategic question.
It may no longer be enough to ask:
“How do people discover my software?”
Increasingly, builders may also need to ask:
“Can AI agents discover, understand and use my software?”
That could be one of the most important consequences of OpenAI DevDay 2026.