OpenAI DevDay 2026: 10 AI Features That Matter for Creators and Small Teams

AI Workflows · 7-minute read

From always-on assistants to shared creative workspaces, OpenAI’s latest announcements offer more ways to develop an idea. Choosing the right responsibility to delegate is still the important part.

OpenAI held DevDay on September 29, 2026, in San Francisco. Its official recap brings together more than twenty announcements. The pattern is clear: OpenAI is expanding the places where AI can work, the tools it can reach and the time it can spend on an assignment.

For a creator, the opportunity might be a better-organised project or a working prototype. For a small business, it might be less manual coordination. The selection below considers those practical uses, with product details checked on October 4, 2026. The workflow examples are suggestions, not results from our own product tests.

1. Dots: ongoing help for projects that keep moving

OpenAI’s dots are persistent assistants powered by GPT-6 Astra. They have a cloud computer, can use connected apps and learn from feedback. Rather than starting every assignment from an empty conversation, a dot can build context around a continuing goal.

A useful creative brief could ask a dot to maintain a research pack, prepare draft launch materials and flag decisions needing attention. Give it examples of your voice, a definition of a good result and clear approval boundaries. That preparation matters more than asking it to “handle everything”.

Access: Dots are rolling out in eligible markets on Pro and Business Premium. Enterprise, Edu and Healthcare workspaces can try an administrator-enabled beta. Review consequential work; persistent assistance still makes mistakes.

2. GPT-6.1 Sol: another way to balance capability and cost

GPT-6.1 Sol improves on GPT-6 Sol across OpenAI’s coding, computer-use and professional-work evaluations. Its standard API prices are US$2 per million input tokens, US$0.10 per million cached input tokens and US$10 per million output tokens. Higher long-context rates and separate cache-write or tool charges can also apply; check the current pricing table for your workload. Those are usage prices, not a fixed price for completing a project.

The practical question is whether it produces an acceptable result on your actual work. Compare a familiar brief, document task or website change across models, including the time spent checking and correcting the output. OpenAI’s benchmark results are useful context, but they cannot establish your cost per successful deliverable.

Access: The launch documentation lists Sol in ChatGPT Work, Codex and the API. It explicitly says the model is not yet available in ordinary Chat.

3. Ultrafast: pay for speed where waiting is expensive

Ultrafast is a premium service tier for faster model execution. GPT-6 Astra support is available in the API; the DevDay announcement also lists access in Work and Codex on Pro 500 and Enterprise. GPT-6.1 Sol Ultrafast was announced as coming soon.

This could matter during interactive prototyping, when each pause interrupts a person’s train of thought. It is less compelling for work that can run unattended. Measure the full task, including tools and network delays, before paying more. Faster token generation does not automatically mean an equally large reduction in project time.

4. ChatGPT Space and Pages: a shared home for the work

ChatGPT Space brings pages and files into a shared workspace. Pages are editable documents where people and ChatGPT can develop text, images, charts and interactive content together.

A small studio could use a project page to keep the approved brief, source material, decisions and review notes together. That makes it easier to see what changed and why. Separate the material collaborators need from private background context: anything placed on a shared page becomes visible to its audience.

Access: Space supports Pro, Business and Enterprise. Page creation and editing are available on web and desktop; mobile creation and editing remain upcoming in the current guide. Collaborative slides and spreadsheets are also described as coming soon.

5. Codex Cloud: continue development beyond one computer

Codex Cloud lets coding tasks run in cloud environments built around a project’s repositories, dependencies and tools. Each task has its own workspace, and work can continue while your computer sleeps. Results can be reviewed from web, mobile or desktop.

For a creative website, start with a bounded change: improve a gallery filter, repair a mobile layout or add a clearly specified interaction. Ask for the changes and test results, then review the outcome before release. Reusable environments make the setup easier to repeat; they do not remove the need to understand what is being shipped.

6. Agents API computer use: browser workflows inside an application

The Agents API entered public beta on September 10. DevDay added a computer-use announcement. The current developer guide describes agents navigating an OpenAI-hosted browser, with applications handling website-access requests, sign-in and session events.

This gives developers another route for a workflow that depends on a website interface. A sensible first experiment would inspect a public site and return a documented usability report. Keep the task narrow, verify the result and design approval points before allowing changes to accounts or business data. This is developer infrastructure, so implementation and testing remain part of the work.

7. Plugin extensions: creative tools closer to the conversation

OpenAI’s plugin extensions give connected products richer interfaces inside ChatGPT. Its examples distinguish Canva’s sidebar design experience, Figma file mentions in the desktop composer, and Adobe file handling for Acrobat and Photoshop.

For creators, the interesting possibility is retaining access to the material being discussed while revising it. Ask which files an integration can read or edit, what permissions it needs and where the finished work will live. The examples support different actions; connecting one product does not grant every capability shown by another.

Access: Composer mentions are desktop-only. The developer guide still lists web extensions for Free and Go as coming soon.

8. MCP events: start a workflow when something changes

MCP events let ChatGPT subscribe to supported updates from a connected service. A matching event can trigger instructions the user has defined. OpenAI’s current integration uses webhooks and supports a draft event specification. Events currently work in Work chats on web, desktop chats with Cloud selected, and with dots.

Consider a review comment arriving on a project: a useful response might be to gather the relevant context and prepare a proposed revision. That is different from the recurring checks covered in our scheduled-tasks explainer. The trigger, source and permitted response should all be explicit. A service must support the relevant events; this is not automatic monitoring of every connected app.

9. Sites with plugins: shared tools with individual permissions

The DevDay update allows supported plugins to be used in ChatGPT Sites. In eligible workspaces, people can approve read-only access to their own connected tools. The Site sees information they already have permission to access. A visitor’s connected-tool information is available while they use the Site, rather than to scheduled background tasks.

A small team could explore a project dashboard that brings approved information into one view. Start with a question the tool should answer, then identify the minimum data required. Check the experience from each user’s perspective instead of assuming everyone can see the builder’s sources. The new feature is the connection to supported tools; Sites itself was already available before DevDay.

10. Decisions API: bounded choices for an application

OpenAI announced Decisions API as a limited preview. It focuses Luna on user-defined questions with finite, predefined answers. Text or image context can inform classification, request routing or the next action in a workflow.

For example, a developer might test whether an incoming request belongs in one of several review queues. The difficult part remains defining useful categories, handling uncertainty and checking mistakes. A limited set of outputs does not guarantee the right choice. OpenAI said a broader release was planned; this article does not assume that release has happened.

What to try first

The most useful starting point is one piece of work you already understand. Decide what a good result looks like, choose a tool with the right access and keep a person responsible for the consequential decisions. Our earlier article on what to delegate remains relevant as these products become more capable.

The AGENTICA IX view: For AGENTICA IX, extending individuality means giving a distinctive idea more room to develop. These tools can help with research, coordination, production and iteration. Creative direction still needs someone who knows what the work is trying to say, and who is willing to change or reject an impressive-looking result.

What to watch next

Watch the rollout details as closely as the demos. Availability, permissions and reliable handoffs will determine which announcements become a useful part of everyday work.

Sources and related reading

Sources and timing: Primary OpenAI launch posts and product documentation are linked beside the relevant claims. DevDay took place on September 29, 2026; availability was checked on October 4, 2026 and can change. Product examples above are proposed uses, not independently measured performance claims.

• OpenAI DevDay 2026 recap
• Introducing dots
• Introducing GPT-6.1 Sol
• OpenAI API pricing
• Ultrafast mode
• ChatGPT Space and Pages
• Codex Cloud
• Introducing the Agents API
• Agents API computer-use guide
• Plugin extensions
• MCP events
• ChatGPT Sites

More from AGENTICA IX

• The Most Valuable Skill in the Astra Era Is Knowing What Not to Delegate
• ChatGPT Scheduled Tasks Can Monitor the Internet While You Are Away