In August, I wrote about completing more work with agents while carrying more responsibility for everything they helped me finish. I still want that ability. I also have to leave room in the day to understand what comes back, and that changes how much work I should start.
Molly Graham returned to that problem in her September 27 conversation with Lenny Rachitsky. Her earlier advice about giving work away assumed another person could take it over. With AI, she argues, the person delegating still carries the oversight and responsibility for the result. The task moves, while much of the mental work stays with the person who assigned it.
That leaves me with a scheduling question. Before I start another piece of work, where will the time come from to read it, question it, and decide whether to use it? An agent can keep working while I am doing something else. My attention will still be needed when it returns, often alongside several other things that have also finished.
the work comes back to someone's calendar
Think about a research report that arrives with a recommendation already written. Even with correct charts and clear sentences, someone still has to decide whether the evidence supports the recommendation strongly enough for the decision at hand. A small change in the wording can turn a useful observation into a claim the research never tested.
Suppose a report recommends moving money into a campaign because response improved. Before acting, I would want to know what improved from what starting point, how many people were involved, and whether the increase would be useful at the scale of the proposed spend. Asking an agent to expand the report could give me more pages without answering any of those questions. A shorter report with those answers would help me decide whether to change the spend.
An illustrative research decision
A report arrives
Response improved. The draft recommends increasing campaign spend.
Check the reasoning
What was the starting point? Is the increase useful at the proposed scale?
Choose what to do
Use the recommendation, ask for more evidence, or stop the work.
The person reviewing that report needs time to follow the reasoning. If the next assignment begins as soon as the draft arrives, the review can become something squeezed between other obligations. Work starts to accumulate at the point where someone has to make a consequential choice.
saving time can send work to the next person
Sophie Alpert described a related problem in Clay's August AI writing policy. Documents that saved their authors time were becoming harder for colleagues to read. Clay allows AI assistance and asks authors to take responsibility for the ideas they share, including whether the document makes good use of the reader's time.
I would apply that same expectation to what I ask an agent to bring back. If I need to choose between two options, I need the difference that affects the choice and enough supporting detail to examine it. When that connection is missing, I should send the work back with a specific question. Asking for more directions can reopen a decision without resolving the reason it stalled.
Exploration can be much broader. I want room for an agent to find something I had not considered, while keeping the result small enough to read and compare with the work already under way. Otherwise, a useful detour can become another assignment before I have decided whether to pursue it.
leave room for the decision before starting
For me, the next improvement is to put the review into the assignment from the beginning. A quick, reversible change can come back with a concise account of what changed and the checks that were run. A recommendation going to a client needs a more careful read, and that time has to come from somewhere in the same working day.
I would rather leave an idea waiting than turn it into another finished-looking document that I have no time to evaluate. That can feel wasteful when an agent is available and ready to work. But starting it creates an obligation for me later, and availability alone gives me very little reason to move it ahead of work that is already close to being useful.
I also want to pay closer attention to work waiting for a decision. How long has it been sitting there? What does the reviewer need to resolve it? Some pieces will be ready with one answer from the agent. Others will need a change in direction, or a decision to stop. Counting them all as completed hides the part of the job that remains.
Agents have made it easier for me to begin work that used to stay out of reach. Before I ask for the next piece, I need to look at what is already waiting and make enough room to finish it.