OpenAI reports a shift from chat to longer delegated agent work

OpenAI published research on 25 June 2026 describing how its employees moved from short chatbot interactions toward delegated, longer-running Codex tasks. The report distinguishes an agent from a single response by its ability to operate for an extended period, use tools and iterate toward an outcome.

OpenAI reports that internal adoption spread beyond engineering into legal, finance, recruiting and support. The evidence comes from OpenAI's own workforce, so it should not be treated as a universal productivity result for every organization.

When several agent tasks can run at the same time, the human role moves toward defining outcomes, supplying context, reviewing progress and deciding what should happen next. That is closer to managing a queue of delegated work than asking a chatbot a sequence of questions.

For a small business, the useful lesson is not to launch many agents immediately. It is to make each delegated task clear enough that success, failure and escalation can be observed.

Begin with one task whose output can be checked, such as preparing a daily enquiry summary, researching a defined account list or drafting follow-up messages for approval. Record the time spent, corrections required and cases that need a person.

Longer autonomy increases the importance of checkpoints. Give the agent a limited tool set, define a stopping condition and require approval before customer commitments, payments, publishing or sensitive data changes.