By Keshavan Thiagaraja
An AI assistant that keeps working after a conversation ends changes how we delegate. For enterprise leaders, the useful question is which responsibilities we can assign, how we will check the work, and who remains accountable.
OpenAI describes dots as ongoing agents that continue work between conversations and bring back results or decisions for review. Access is rolling out gradually to eligible accounts.
My view is that the business opportunity lies in reducing the effort required to keep work moving. Research, follow-ups and document preparation consume time because people must repeatedly gather context and decide what happens next. An agent that maintains continuity could help. Its value still needs to be demonstrated in the workflow where it will operate.
What OpenAI has introduced
According to official OpenAI documentation, a dot runs on GPT-6 Astra and has its own cloud computer and browser. It can research, analyse data, prepare documents and build software. Its cloud work can continue when the user's computer is off. [2]
This needs a careful distinction. A cloud computer, a connection to an app, and access to a local computer are separate capabilities. Connecting a messaging channel does not automatically grant access to other services. Each connection has its own permissions and requirements. [2]
For a CTO, that distinction matters when designing access. For a CEO, it matters when agreeing what the agent should be responsible for.
A practical enterprise example
Consider a hypothetical procurement team preparing a supplier renewal. The information sits across a contract, usage reports and several emails. A manager currently assembles it before every discussion.
A sensible pilot would ask a dot to maintain a renewal brief using approved sources, identify missing information and prepare questions for the manager. The team would review the first result before expanding the responsibility.
The instruction should define the output and its limits: keep the brief current, show the sources behind important figures, highlight unresolved assumptions, and prepare correspondence for review. The manager retains responsibility for the commercial decision.
This is an illustrative workflow, not a claim that OpenAI has validated this particular use case or delivered a measured saving.
Compare systems by the work they must perform
I would compare three approaches before choosing a platform.
A conversational assistant helps someone complete a task during an exchange. An ongoing agent carries a responsibility across exchanges. The choice depends on whether continuity solves a real problem.
A predefined workflow follows explicit steps. An agent can propose how to respond when information changes. Stable, repetitive processes may benefit from conventional automation; work with changing inputs may justify an agent pilot.
A demonstration shows what happened once. An operational evaluation asks whether the system produces acceptable results repeatedly, including when a source is missing or contradictory. Enterprise purchasing should be informed by the latter.
What CEOs should do
Choose one bounded responsibility with a named business owner. A supplier brief, an internal project update or preparation for a recurring meeting offers a clearer starting point than a broad instruction to improve productivity.
Measure the current process before the pilot. Record staff time, turnaround time and the corrections needed to produce an acceptable output. During the pilot, include review time and rework in the comparison.
Ask whether the result helps the team make a better decision or meet a deadline. More generated documents do not necessarily mean more business value.
Agree the conditions for expansion in advance. The business owner should be able to explain which results are acceptable and which failures would stop the pilot.
What CTOs should establish
Define which sources the agent may use and which actions require a person. Begin with the smallest useful access scope. Test with representative data and deliberately incomplete inputs before extending access.
Require evidence for important outputs. A renewal brief should make it possible to trace a quoted amount to the source and distinguish a current fact from an assumption.
Evaluate the whole workflow. Check retrieval quality, calculations, document quality, access boundaries and recovery after a failed step. A correct paragraph cannot compensate for the wrong source file.
Set a practical interruption and recovery procedure. OpenAI notes that stopping work does not undo completed actions, and that delegated or recurring tasks may need separate controls. Teams should understand those controls before assigning ongoing responsibilities. [2]
Monitor the cost of the work performed, including human review. Avoid extrapolating a successful small pilot into a company-wide saving without further evidence.
The decision ahead
Dots make ongoing delegation a concrete product proposition. Whether that proposition improves an enterprise depends on the responsibility assigned, the information available and the quality of the review.
I would start with a task whose result can be checked, learn from the corrections, and expand only when the evidence supports it. That gives leaders a basis for investment and gives employees a clear role in shaping how the tool is used.
Which ongoing responsibility in your organisation would you assign to an AI agent first, and what evidence would you require before expanding its authority?
Think mathematically. Engineer intelligently. Lead strategically. Build responsibly. Scale globally. Create real business value through artificial intelligence.
#OpenAI #Dots #EnterpriseAI #AIStrategy #AILeadership #ResponsibleAI
Originally published on LinkedIn by Keshavan Thiagaraja. Read the original on LinkedIn.
