The Biggest Announcements From OpenAI DevDay: Dots, Agents and a $500 Plan

OpenAI’s DevDay 2026 keynote showed a company trying to move ChatGPT from a conversational interface into a persistent work layer for businesses.

The announcements centered on dots, OpenAI’s always-on agents that can take on ongoing responsibilities, along with a shared ChatGPT Space, new agent-building APIs, a Decisions API for constrained business tasks and higher-capacity pricing for power users.

Dots, the AI Agents That Never Log Off

A big focal point for the event was OpenAI’s introduction of dots — always-on AI agents that can learn what matters to users and take on responsibilities on their behalf.

While dots are an interesting step away from conversational AI chatbots, they also bring new control problems.

A chatbot can be governed at the level of access, prompts and data use. An always-on agent requires more attention to permissions, task boundaries, approvals and monitoring, because it can act across tools or continue work after the user has moved on.

OpenAI appears to recognize that risk: Enterprise, Edu and Healthcare users can try the dots beta only when a workspace admin enables it. The feature is off by default.

At launch, each user begins with one primary dot. OpenAI plans to let users add more in the future. Through ChatGPT’s plugin ecosystem, a dot can connect to more than 4,000 apps. Users can communicate with it by text or voice in ChatGPT, with access also available through Slack and Microsoft Teams.

ChatGPT Space Gives Humans and Dots a Shared Desk

Dots are more important for enterprise users when they go from personal assistance to shared work.

OpenAI’s ChatGPT Space is designed as a place where teammates, ChatGPT and dots can work from the same body of knowledge, making the agents part of the team environment rather than separate tools each employee uses on their own.

In practice, that could make ChatGPT Space a shared workspace for projects, documents, meeting notes, research, planning and recurring tasks.

A dot that understands the goals and context of a space (like a product launch date and launch plans, for instance) could organize information, update materials or keep work moving across people and tools.

Shared knowledge also creates shared risk, however. Companies will need to understand what a dot can access, whose instructions take priority, how changes are tracked and when a person must approve next steps. If ChatGPT Space becomes a place where teams and agents work together, governance will need to operate at the workspace level, not just the individual user level.

Agents API Lets Developers Build While OpenAI Handles the Plumbing

OpenAI also extended its agent strategy to developers through the Agents API. The company said the API now supports computer use, which allows AI agents to interact with software to complete tasks.

The Agents API also brings multi-agent capabilities, tool search, tool calling and context compaction into applications, with OpenAI running the underlying infrastructure.

Building enterprise agents requires more than a capable model, however. Companies also need orchestration, tools, memory, permissions, monitoring and ways to recover when a task fails. By managing more of that infrastructure, OpenAI is offering developers a faster path to building agents that can work inside business applications and existing processes.

The trade-off is platform dependence. If companies use the Agents API as the foundation for agentic workflows, OpenAI becomes part of the execution layer for business tasks. While that could reduce the burden on internal engineering teams, it also means another vendor-controlled system companies must evaluate for reliability, observability, security and long-term cost.

Decisions API Gives AI a Menu, Not a Blank Check

OpenAI’s Decisions API may offer a more manageable entry point for companies still weighing how much authority to give AI.

OpenAI says the API focuses GPT-5.6 Luna’s intelligence on predefined questions with a finite set of answers. Developers provide text or images as context and receive answers for tasks like content classification, request routing or next-step actions. It launched in limited preview, with broader availability planned in the coming days.

Within an agent workflow, the Decisions API could select next steps from an approved list — like retrieving more information or escalating a case to a human reviewer. Essentially, businesses can define the available choices and connect each answer to an established process.

Dots are designed to take on ongoing responsibilities, which makes their work hard to evaluate against a single right answer. The Decisions API works the other way: because the possible answers are defined in advance, teams can test performance against known outcomes, log decisions and determine which actions each result may trigger. The constraint applies to the output, however, and doesn't guarantee the model will choose correctly. Enterprises will still need fallback rules, escalation paths and controls over any downstream action.

Pro 500 Puts AI Agents in the Fast Lane — for $500 a Month

OpenAI also introduced a higher-priced plan aimed at users who need more speed and capacity for agent work.

The new Pro 500 plan, priced at $500 per month, provides what OpenAI calls its highest usage allowance — equal to 25x the ChatGPT Plus allowance — and includes Ultrafast, the company’s premium speed tier.

OpenAI says that the premium speed tier can generate tokens up to 8x faster in Codex and up to 6x faster through the API. GPT-6 Astra Ultrafast is available through the API and on Pro 500 and Enterprise plans.

Speed could matter more as agents attempt multi-step work. A task that requires repeated model calls, tool use and coordination among agents can accumulate delays at each stage. Faster generation could make those workflows more practical for settings where time is of the essence — like customer support or software development.

Still, the advertised gains describe token generation, and that doesn’t necessarily translate into an equivalent reduction in task times.

For enterprises, the larger issue is how to match performance and cost to the work. Premium capacity might be worth the cost if the use case justifies it — like a workflow that directly affects customer experiences or worker productivity.

As agent use grows, businesses will need to manage model speed, usage allowances and workload priority as part of the same operating model they use for permissions and oversight.

Private Intelligence and AWS: OpenAI Makes Its Trust Pitch

OpenAI paired its agent announcements with new options for protecting data and controlling deployment.

Its Private Intelligence offering has two parts:

1. Private Safety Processing runs automated safety reviews without giving OpenAI staff access to the underlying content.
2. Private Inference, a confidential-computing option, is expected this fall.

OpenAI also worked with Amazon on Bedrock Managed Agents, allowing enterprises to run OpenAI-powered agents entirely within AWS and connect them to existing AWS resources.

The Bottom Line: More Capable Agents Need Stronger Guardrails

DevDay 2026 made clear that OpenAI wants its agents working inside everyday business operations, not alongside them.

Agents that keep working between conversations, collaborate in shared spaces, use software and make defined decisions could make core processes more productive. But the deeper agents are embedded, the more weak oversight costs. Enterprise leaders should treat agent adoption as a change to how the business operates, with governance built up at the same pace as capability. 

Have a tip to share with our editorial team? Drop us a line: