OpenAI dots and the next decision enterprise leaders must make
An AI assistant that keeps working after the conversation ends changes how we hand over responsibilities. These are the questions enterprise leaders must consider: which responsibilities can be delegated to it? How do we verify its work? Who accepts final accountability?
OpenAI describes dots as ongoing agents that keep working between conversations and bring results, or matters needing a decision, up for review. The feature is being rolled out gradually to eligible accounts.
In my view, the business opportunity lies in reducing the effort needed to keep work moving forward. Research, follow-ups and document preparation take time because each round a human must gather the background and decide what to do next. An AI agent that preserves the continuity of work can help - but its benefit must be proven in the workflow where it is used.
According to OpenAI's official documentation, a dot runs on the GPT-6 Astra model and has its own cloud computer and web browser. It can do research, analyse data, prepare documents and build software, and work in its cloud environment continues even when the user's computer is off.
One important distinction must be understood. A cloud computer, a connection to an app, and access to the user's own computer are separate capabilities. Connecting a messaging channel does not automatically grant access to other services; each connection has its own permissions and requirements. A CTO must watch this when designing access, and a CEO when deciding which responsibilities to delegate.
Imagine a company's procurement team preparing to renew a supplier contract. The information needed is scattered across a contract, usage reports and many emails. In a suitable pilot, a dot could be asked to keep a renewal brief continuously updated using approved sources, spot missing information and prepare questions for the manager. The team should review its first output before expanding its responsibilities. Accountability for the commercial decision stays with the manager. This is an illustrative workflow - OpenAI is not said to have tested this use case or provided a measured saving.
Imagine a company's procurement team preparing to renew a supplier contract, with the information scattered across a contract, usage reports and many emails. In a suitable pilot, a dot could keep a renewal brief continuously updated from approved sources, spot missing information and prepare questions for the manager - who keeps accountability for the commercial decision. This is an illustrative workflow; OpenAI is not said to have tested this use case or provided a measured saving.
Before choosing a platform, I would compare three approaches. A conversational assistant helps within one conversation; an ongoing agent carries a responsibility across many. A predefined workflow follows fixed steps, and conventional automation may serve stable, repeated processes better. A demonstration shows what happened once; an operational evaluation tests whether the system keeps delivering acceptable results when information is missing or contradictory. The purchasing decision should rest on that evaluation.
CEOs should pick one responsibility with clear boundaries, appoint an accountable business owner, measure the current process before the pilot - staff time, turnaround time, corrections - and count review time and rework in the comparison. CTOs should start with the minimum access scope, require sources for important outputs, evaluate the whole end-to-end workflow, and set up an interruption and recovery procedure: OpenAI notes that stopping work does not undo actions already completed.
Dots present ongoing delegation as a practical product opportunity. I would start with a job that has a verifiable outcome, learn from the corrections it needs, and expand only when the evidence supports it. Which responsibility in your organisation would you delegate first?
