Nate B Jones distinguishes visible agent activity from outcomes that are worth the human effort needed to review and actCost per completed task measures the total AI, tool, infrastructure, retry, and repair expense for each verified useful outcome. on them. A report, proposal or estimate can be cheap to generate yet still impose substantial work on its intended readers.
He recommends using parallel agents to investigate different explanationsAI agent delegation assigns a bounded task, context, authority, and expected result from one participant to an agent or another agent. before committing to solutions. Experiments can be valuable even when their results are discarded, and some questions still require direct contact with customers or other real-world evidence.
Before turning an assignment into a recurring taskWorkflow automation uses software or AI to complete a connected sequence of routine steps with less manual effort., Jones suggests deciding when it should stop. A scheduled update can continue consuming resources and attention after the decision it supported has already been made.
Finally, he describes giving a coordinating agent an attention budget: it should understand existing commitments, filter interruptions and present choices that matter to its human manager.
Watch on YouTube
