At the end of Pratyush Sinha’s TechDays talk on coordinating AI agents, someone in the audience asked an unexpectedly good question: what had he learned from managing agents that he would now apply to managing humans?
The answer did not travel very far, but the question did. Building agentic systems forces people to specify goals, handoffs, decision rights and failure routes that human organisations often leave vague. Humans keep those organisations working by interpreting, negotiating and quietly repairing the gaps.
A vague brief transfers work
When an AI agent receives an underspecified task, the weakness becomes visible quickly. It makes a plausible assumption, produces the wrong kind of result or continues in a direction no one intended. Designers then improve the prompt, add context, define acceptance criteria or create an escalation route.
Human managers also give vague briefs, but the consequences can be harder to see. An experienced colleague infers the purpose, notices what was omitted and works out which interpretation will survive contact with the organisation. The task appears to have been delegated successfully because the recipient supplied the missing management.
What looks like human flexibility is often invisible coordination work. It has value, but it also consumes time and distributes advantage unevenly. People who already know the organisation’s unwritten rules can repair a poor brief more easily than newcomers, outsiders or anyone who interprets language more literally.
Communication contracts make the handoff visible
Agent architectures often use something resembling a communication contract. One component must produce information in an agreed form so that another component can use it. The contract defines what is passed, what counts as complete and what should happen when the input is inadequate.
Human work rarely needs a rigid schema for every conversation. Still, the underlying questions are useful. What does the next person need in order to act? Which decisions have already been made? What remains open? Who can resolve an ambiguity? What should happen when the work does not meet the agreed standard?
A good handoff is not simply a clear message. It connects information to authority and a route for recovery.
Clear interfaces can still lose information
There is a limit to the analogy. Formalising communication can make work easier to coordinate, but it can also strip away information that does not fit the contract. A structured field captures the expected answer. It may not capture a doubt, an emerging pattern or the sense that the question itself is wrong.
Agent systems make this trade-off unusually visible because each boundary has to be designed. Human organisations have the same problem, even when their interfaces are called forms, templates, reporting lines or meeting agendas.
The aim is not to specify everything. It is to decide where consistency is useful and where the recipient needs room to question the frame.
Humans are not agents with worse compliance
Lessons from agent coordination can easily slide into a bleak managerial fantasy: define the contract, monitor performance and repair deviations. That misses what humans contribute.
People do not only execute a goal. They can notice that the goal is unwise, that the evaluation criterion rewards the wrong behaviour or that the task conflicts with another responsibility. They bring histories, loyalties, values and knowledge that were never included in the brief. Sometimes the failure to follow an instruction is the most intelligent part of the system.
Good human management therefore cannot be reduced to better specification. Clarity should help people exercise judgement, not remove their permission to use it.
Agent design exposes an old organisational dependency
The useful lesson is not that people should be managed like agents. It is that designing for agents reveals how much ordinary management relies on humans absorbing ambiguity.
Once that repair work is visible, several questions become harder to ignore. Are goals actually shared, or merely stated? Do people know which decisions they own? Can they return an inadequate brief without being treated as obstructive? Is disagreement part of the system, or evidence that someone has failed to align?
Agentic systems require explicit coordination because the agents cannot be trusted to reconstruct the organisation around the task. Humans often can, until they cannot, or until the effort becomes exhausting.
Perhaps the most useful thing managing agents can reveal about managing humans is not how to control people more precisely. It is how much competent people have been doing to compensate for underspecified work all along.