🔍 Read the full analysis: What Developers Can Learn From OpenAI Dev Day 2026 on ThorstenMeyerAI.com
Get everyday essentials delivered free — and shop member deals
- Fast, free delivery on millions of items
- Access to Prime Big Deal Days deals on October 6–7
- Prime Video, Amazon Music and more included
TL;DR
At DevDay on Sept. 29, OpenAI announced more than 20 products and features, including persistent Dots agents, plugin extensions, computer use in the Agents API and the limited-preview Decisions API. The announcements point to a platform strategy built around agents that can use software and services, though access, reliability, costs and the commercial terms for several offerings remain unsettled.
OpenAI announced more than 20 products and features at its Sept. 29 DevDay, putting persistent AI agents, software integrations and developer APIs at the center of its platform strategy. For builders, the practical shift is toward systems that can choose tools and carry out tasks, not only generate answers; several new capabilities remain in preview or have limited availability.
The most visible announcement was Dots, persistent agents powered by GPT-6 Astra that OpenAI says can work toward goals over time. Each Dot gets a separate cloud computer and browser and can connect to more than 4,000 apps through plugins. OpenAI says users can reach Dots through ChatGPT on desktop, web and mobile, as well as Slack and Teams. The company’s examples include monitoring customer feedback, updating a sales proposal as requirements change and rerunning an analysis when new data arrives. Those are company-described use cases, not independently verified performance results.
OpenAI also added computer use to its Agents API, allowing agents to operate software through its interface, with OpenAI running the Codex harness. Its new Decisions API is designed for bounded choices: developers specify possible answers, provide text or image context, and receive a selection for tasks such as classification or routing. The API is in limited preview. OpenAI also introduced GPT-6.1 Sol, which it describes as near-Astra intelligence at one-fifth Astra’s token prices.
Plugins are becoming a shared distribution layer across ChatGPT, Codex and Dots. Announced changes include extensions that appear in ChatGPT, interactive panels, improved discovery and a revised creator and submission flow. OpenAI says plugins can connect external services through MCP servers; support for MCP events could let an app trigger an automation when something changes in a connected service. Sites can host plugins for teams, and an OpenAI Marketplace will let eligible enterprise customers spend part of their OpenAI commitment with launch partners.
OpenAI DevDay 2026 at a glance
The five launches that matter, plus Sol
The rest of the slate
Collaboration
ChatGPT SpacelivePagesliveCollaborative slidesnext few weeksTeams and Team Taskslive@ChatGPT in Slack and TeamsliveMeetings pluginbeta, macOSShareable profilesliveCodex
Codex CloudliveCLI refreshvoice, /agentsCode ReviewliveCodex Security CloudlivePlans, enterprise, speed
Pro 500$500, 25× PlusUltrafastAstra now, Sol soonPrivate IntelligencePrivate Inference this fallBedrock Managed Agentsruns in AWSOpenAI Marketplace32 partners, applyBuilding for Agents and App Discovery
For developers, the announcements raise two linked questions: what work can an agent finish, and how will it find the service that can do it? If a user describes an outcome and ChatGPT chooses a plugin to perform the task, discovery may happen inside an agent conversation rather than through a user searching for a vendor by name. OpenAI’s framing suggests businesses may need to make their services legible to agents and track whether users activate and complete a task, alongside conventional measures such as visits or downloads.
The APIs also offer different levels of control. The Decisions API limits a model’s answer to choices defined by the developer, which could make it suitable for narrow routing decisions. Computer use gives an agent a broader way to interact with existing software, but also places more weight on permissions, testing and error handling. Teams evaluating these tools should match the level of autonomy to the consequences of a mistake and measure task completion against their current workflow.
Cost and access will shape whether the platform is practical. OpenAI says GPT-6.1 Sol approaches Astra’s intelligence at a fifth of the token price, a claim developers can test against their own tasks and budgets. The company also says the first Dot is included in certain plans, while work started in Codex or ChatGPT Work counts toward usage limits. These terms make workload design and usage monitoring part of the product decision.
From Chat Interface to Platform
OpenAI held DevDay at Fort Mason in San Francisco on Sept. 29, 2026, with a keynote and technical sessions for developers. Its event page described the gathering as a venue for product demonstrations, workshops and discussion with the teams building OpenAI tools. At the keynote, the company presented ChatGPT as a shared surface for people and agents and cited 1.2 billion weekly users. That audience figure is OpenAI’s own statement.
The announcements extend earlier developer-facing work on models and APIs. The Agents API changes aim to let software agents interact with user interfaces, while plugin updates give those agents routes into connected services. The combination matters because an agent’s usefulness depends not just on the model generating a plan but also on access to the tools, information and permissions needed to carry it out.
OpenAI described Dots as its consumer agent product and outlined specialist Dots for enterprise pilots in areas such as procurement, invoice processing, customer support and commercial contracting. It said specialist agents would have their own identities, credentials and IT-provisioned hardware, with a Microsoft Agent 365 integration planned for governance. These pilots indicate the company’s enterprise direction; they do not establish how widely the tools will be deployed or how well they will perform.
Access, Reliability and Usage Costs
Several practical details remain open. The Decisions API is in limited preview, and the source material does not establish when it will be broadly available or provide a full account of its supported workloads and service limits. Availability for Dots also differs by plan and market: OpenAI says Pro and Business Premium users in eligible markets can access them, while Enterprise, Edu and Healthcare customers can use a beta when an administrator enables it.
OpenAI says Dots can make mistakes and describes controls including custom rules, approval requirements, an activity view and the ability to pause or stop an agent. The company says background proactive research is restricted to read-only tools and that sensitive actions, such as changing a password, remain with the user. It is not yet clear how these controls will perform across every connected app, how often users will need to intervene, or how costs will change as agents run longer tasks or users add capacity.
The company’s claims about model quality, productivity and user reach should be treated as OpenAI’s claims unless independently measured. Developers also need more detail to judge how plugin selection will work in practice, what analytics will be available and what commercial conditions apply to Marketplace distribution.
Test Workflows Before Scaling
The next step for developers is to test the tools as they become available: compare GPT-6.1 Sol with current models on representative tasks, try the Decisions API on tightly scoped choices, and evaluate computer use in a controlled environment. Plugin builders can review the new submission flow and assess whether extensions, MCP connections or event-triggered automations fit their existing service.
Enterprise teams will need to define which actions an agent may take independently, which require approval and how activity will be monitored. OpenAI says it plans specialist Dot pilots and a Microsoft Agent 365 integration; it has not provided a launch timetable in the supplied announcement material. Availability and product terms will determine when teams can move from prototypes to broader use.
The clearest lesson from DevDay is to start with a measurable task, not with an agent label. Record how long the task takes, the rate of successful completion, intervention frequency and total model and tool costs. Those results will show whether an agent or plugin improves a real workflow as access expands and details become clearer.
Key Questions
What was the main developer news at OpenAI DevDay 2026?
OpenAI announced more than 20 products and features, including Dots, computer use in the Agents API, plugin changes, the Decisions API and GPT-6.1 Sol.
What does the Decisions API do?
It lets a developer define a finite set of possible answers and provide text or image context for the model to choose from. OpenAI says it is in limited preview.
How do Dots connect to other software?
OpenAI says Dots have a cloud computer and browser and can reach more than 4,000 apps through plugins. Access depends on plan, market and, for some organizational plans, administrator settings.
What should developers evaluate first?
Start with a specific, repeatable task. Compare completion quality, time, required human intervention and total usage costs against the existing workflow before expanding an agent’s permissions.
Source: ThorstenMeyerAI.com
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
