After OpenAI DevDay 2026: work between apps is the next big question

The most consequential DevDay announcement may be the connection between the products. A message arrives, an agent starts work, colleagues edit the result, and the same subscription can help fund work in another app. OpenAI is trying to make ChatGPT the place where a job keeps moving. That is Hammer's reading of the September 29 announcements.

Tuesday's keynote report covered dots, Space, GPT-6.1 Sol, Ultrafast and browser use. This next-day synthesis adds the mechanics behind shared tasks, event-driven plugins, reusable cloud environments and subscription sharing, plus a more precise answer about dots in Sweden.

Coverage: DevDay took place on September 29 in San Francisco. The closing session was scheduled to start at 01:00 Swedish time on September 30. At our morning recording check, we could verify the keynote and official product demos, but had not located and verified technical-breakout or closing-session replays. OpenAI says other sessions will be posted shortly after the event. This analysis uses the available official announcements, documentation and recordings; Hammer has not tested the products.

Source: OpenAI DevDay schedule and recording policy.

The practical shift: work can start when something happens

MCP, the Model Context Protocol, connects AI applications to external tools and data. With MCP Events, a connected application can notify ChatGPT of a new message, document comment or status change. The user specifies what to watch and what should happen next.

The developer documentation makes this concrete: a plugin's server advertises events, ChatGPT subscribes, and the server sends matching updates to a verified callback address. ChatGPT then follows the instructions in the subscribed conversation. The integration requires MCP 2.0, protocol version 2026-07-28, and currently supports webhook delivery from the draft Events specification. An existing tool connection needs event support implemented before it can do this.

Hammer's assessment: this is commercially interesting for work that loses time waiting for someone to notice a change. A customer comments on a proposal; an automation could gather the relevant material and prepare a revision. The trigger supplies the reason to start. The connected tools and instructions determine the work that follows.

Source: MCP Events developer documentation.

Team Tasks give recurring work a shared owner

Team Tasks run in the cloud on a schedule or a supported event. OpenAI lists them for Business and Enterprise. They use the team's service account and configured app connections, and consume workspace credits. Authorized colleagues can change the instructions and manage the task.

That extends Tuesday's Space story. Space and Pages provide shared material; Team Tasks provide recurring execution. The documentation says tasks are designed to persist after their creator leaves, although ownership and required connections still need attention. Tasks require their own instructions and sources because they do not inherit the creator's personal chat history or memories.

Consider a weekly customer update. The valuable change would be a team-owned preparation process, with a shared result and colleagues who can maintain it. OpenAI also announced @ChatGPT in Slack and Microsoft Teams, where teammates can contribute to a conversation without each needing an individual ChatGPT license. Account access and the workplace setup still determine which tools can be used.

Sources: Team Tasks setup and billing and OpenAI's official DevDay recap.

Codex Cloud makes the project setup reusable

The official Codex Cloud demo adds a useful detail beyond the keynote headlines. Around 01:37, OpenAI demonstrates environment setup: select repositories, let Codex investigate their requirements, and configure networking, environment variables and secrets. Around 02:06, the environment is published as a reusable starting point for separate tasks and teammates.

The documentation confirms that each task gets its own workspace and can continue while the user's computer sleeps. For a developer or an agency, the opportunity is less repeated setup and easier continuation across devices. The time saving remains something to measure in the actual project.

Code Review and Codex Security Cloud broaden that approach: OpenAI describes cloud-based reviews and repository scans that can investigate findings and prepare fixes while the laptop is closed. These are announced capabilities, with plan-specific access, rather than results of a Hammer evaluation.

Sources: Codex Cloud documentation, official Codex Cloud demo and DevDay recap.

ChatGPT becomes a distribution channel and a shared usage budget

Plugin extensions give software a working interface inside ChatGPT. Sites can now host supported plugins, with colleagues using their own connected data and permissions. Together, those announcements suggest a distribution choice for software companies: let customers reach a useful part of the product where they already discuss the work.

Sign in with ChatGPT adds a separate commercial change. Plus and Pro users can choose to spend eligible plan usage in participating apps. The launch names partners including Notion, Vercel and Cognition's Devin. This requires support from the app; commercial integration is initially limited to approved partners.

The important billing detail is in the Help Center: app requests draw on the same included ChatGPT Work and Codex usage. A weekly app limit caps consumption within that shared allowance. The partner can still charge for its own subscription, infrastructure or features, and spending credits beyond the included allowance requires a separate opt-in.

For procurement, compare the whole arrangement: the partner's fee, the shared usage it consumes and the work it replaces. Enterprise buyers have another route through OpenAI Marketplace, where eligible customers can apply part of an existing OpenAI commitment to approved partner software.

Sources: ChatGPT subscription usage in partner apps and OpenAI's DevDay recap.

What Swedish buyers can use, and what remains a preview

The morning documentation sharpens several launch-day distinctions:

• Dots: the Pro rollout excludes the European Economic Area, Switzerland and the UK. Business Premium is listed across supported ChatGPT regions; Enterprise, Edu and Healthcare have an admin-enabled beta. Rollout can take several days. A Swedish Pro subscription therefore does not establish access to dots.
• Decisions API: a limited preview for answering user-defined questions with predefined answers, using Luna. OpenAI names classification, routing and choosing an agent's next action as uses. Wider release is planned in the coming days.
• Private Intelligence: Zero Data Retention with Private Safety Processing uses customer-controlled storage for encrypted safety records and automated review. The documentation requires customers to retain those records for at least 30 days. The separate Private Inference preview is planned for the fall.
• Agents API: its current documentation still specifies US-only data residency and no Zero Data Retention support, including with a self-hosted sandbox. Buyers need the terms for that particular service.
• Still upcoming: collaborative slides and GPT-6.1 Sol Ultrafast. Keep these separate from features your account can use today.

Sources: Dots plans and regional availability, Private Safety Processing, Agents API terms and DevDay recap.

The business question after DevDay

Sol's capability claims and Ultrafast's speed were the attention-grabbers on Tuesday. The broader announcement set points to a contest over where work starts, where it runs and where people finish it together.

A useful first experiment is one recurring deliverable with an identifiable trigger: a changed proposal, an incoming support issue or a weekly customer update. Record the current preparation time, then measure the complete alternative, including corrections, review and total cost. Choose the available product that removes the most repeated coordination from that job.

For an organization that already pays for several workplace tools, that result is more useful than counting new features. It tells you whether another AI subscription earns a place in the workflow.

Official recordings available for this follow-up

Watch the full opening keynote for the launch overview and the Codex Cloud product demo for the reusable-environment workflow. Both were published by OpenAI on September 29. The demo is a separate product video, not a technical-session replay.

Breakout and closing-session recordings were not located and verified during our September 30 morning check. Their contents are outside this article's evidence base. This is a next-day analysis of the event's published material, not a second conference day.