Apps a Dots agent can plug into: More than 4,000 (Source: InfoWorld (October 1, 2026))
Delegation starts with access
Sam Altman says you can hand work to Dots "the way you would to a high agency engineer or a chief of staff." The comparison is useful, though perhaps not for the reason OpenAI intends. A new chief of staff is rarely given a task list alone. They also get inboxes, calendars, systems and sign-off rights.
With an agent, that permission list is the job description. This is WebPulse's view, not the report's. But the launch makes the question concrete, because Dots is already available to paying customers.
What OpenAI announced
InfoWorld reported that OpenAI introduced Dots at its Dev Day this week. Each Dots agent has its own cloud computer. Users reach it in ChatGPT, Slack or Teams. It connects to more than 4,000 apps, within boundaries that humans set.
Dots is available to ChatGPT Pro, Business Premium and Enterprise users. OpenAI is also previewing specialist Dots for enterprises, aimed at accounting, marketing and legal work. It is working with Microsoft to bring Dots into Agent 365.
How it works
Altman's own example is migrating an app off a legacy API, the interface other software calls. That is hard because the API may be used all over a code base, and changes can break things. He said Dots can trace the dependencies, decide what must change, run tests and suggest next steps.
His other examples ranged from engineering to finance. An agent could monitor incoming bug reports and propose fixes, draft pull requests, or kick off next year's budget planning cycle. According to Altman, it learns over time without constant instruction.
A separate Agents API supports computer use. It supplies harnesses (the code that runs an agent through its steps), hosting, memory and multi-agent controls. In plain terms, Dots is an always-on agent that acts on a computer and plugs into many apps. Each connected app is a place where it can read or change something.
Cheaper agents mean more actions
OpenAI also released GPT-6.1 Sol. Altman's claim is that Sol gets close to Astra's capability while charging a fifth of the price. Cached inputs, meaning context the model has already been sent and can reuse, cost $0.10 per million tokens. Tokens are the small chunks of text a model reads.
Analyst Carmi Levy said Sol lowers the cost curve for firms deploying agents at scale. When each run costs less, organisations can run more of them. The cost of checking an agent's work is a separate matter, and the launch coverage does not address it.
What the report leaves open
The report does not explain how Dots boundaries are enforced or logged. It does not say whether limits apply per task, per user or per app.
It also describes Dots as powered by GPT-6 Astra. The same report adds that OpenAI canceled Astra's general availability this week, after the model failed internal safety and alignment tests. The report does not reconcile the two points or say what the tests found.
Altman said some things should not be automated. The same launch includes Codex Security Cloud, which OpenAI says helps defenders find vulnerabilities and prepare fixes.
Questions for your team
Who in your company can switch Dots on today, and which apps can it reach? Are limits set per task or per account? Can you see a record of what an agent did and why? Which work have you decided not to automate?
One more question is WebPulse's own. The report does not say who reviews what Dots drafts. Before an agent drafts pull requests or opens a budget cycle, decide who checks that work and how.
Agents are getting cheaper to run. Deciding what they may touch is still a human job.
Produced by the WebPulse Newsroom with AI assistance from the original reporting credited below, and checked against that source by our editorial review. How we use AI.
Original reporting: InfoWorld.
