When Your Agent Team Outruns Your Attention

In my past life as a consultant, client emergencies often blindsided us. A client would suddenly need a strategy deck or proposal for a high-stakes meeting, and the week would reorganize itself around that demand. I would cancel weekend plans. The team would push through all-nighters to build a PowerPoint we believed would move a real decision. Monday morning, still riding the adrenaline of hitting the deadline, I would learn the ending: the client had no time on the calendar to review it. Not that week, sometimes not that month. The deck was finished. The decision it existed to serve had not even been scheduled. The defeat was always the same. Someone had created urgency without reserving the attention to act on the result.

Last week, I recreated the same failure with my own agent team in Paperclip. This time, I was the client. I had built the team for leverage, and the leverage worked. Completed work piled up in review faster than anything reached production: drafts, plans, pull requests, each of them finished, each of them waiting on me. And every unreviewed piece was a judgment call I had not made yet. Merge or publish it. Restart from the original request. Revise the plan. Redirect the execution. Abandon it. Choosing well meant reloading the request, the plan, and the bar the work had to clear, one artifact at a time. The backlog was a ledger of unmade decisions, growing on exactly the schedule my team was built to accelerate.

The coordination tax

The pileup was the coordination tax: the cost every serious operation pays to keep goals, plans, feedback, and standards moving together. I had multiplied production and routed the entire bill to myself.

I knew the tax well because I spent seven years at Asana, a work management platform, helping organizations transform how they worked. Asana was built on a work graph: you could see who was doing what, by when, across the entire company. Goals, plans, tasks, owners, and retrospectives lived in that shared place. Coordination created clarity across the organization, and that clarity let teams move faster and sustain their velocity. But the clarity was never free. Leaders had to mandate the change. Everyone had to keep their tasks and projects up to date. Humans paid a recurring administrative cost to maintain that clarity. We accepted the tax because clarity beats chaos.

Make every review compound

What is changing in my operation is who pays. I now treat every review I give as something the system should keep. Project memory captures the feedback from each work cycle and distills it into lessons. The lessons that prove durable move into a company-level LLM wiki that later agents and workflows read before they start. And dedicated critics now own narrow quality gates, applying standards I made explicit before work ever reaches me. When I catch a mistake now, the correction outlives the artifact in front of me. I see feedback I gave weeks ago honored in work I never touched. The next draft arrives already knowing what the last review taught, and my attention compounds instead of evaporating.

Raise the floor, then the ceiling

Prose is where this stopped being theory. For weeks I ran anti-slop passes and wordsmithing rounds over LLM drafts, and the writing never became something I enjoyed reading. I could tell it missed my bar. I could not explain why, and I could not fix it consistently myself. So I studied the craft I was trying to demand, working through Good Writing by Neal Allen and Anne Lamott and Writing Without Bullshit by Josh Bernoff. I translated what I learned into explicit editing principles, encoded them, and hired Aster, a specialist agent whose entire job is the final sentence-level pass. The order mattered more than the hire. Before an agent could carry my quality bar, I had to learn to articulate it.

The result changed my ambition. Aster can take a draft I cannot force myself to finish reading and return one I enjoy reviewing and editing. I built all of this to raise the floor, to make the work reliably good enough. Now the floor mostly holds without me, and my attention is climbing toward the constraints that cap the operation: creative direction, distribution, measurement, the design of the content system itself. None of this is blind trust. It rests on memory, explicit craft rules, and standards applied before work reaches me. Aster has not eliminated my editing. It has changed what my editing is for. The evidence is young, and content is only the arena where I learned it. But I see the same shift available anywhere an agent team does serious work.

I think about those consulting weekends differently now. The failure was never a shortage of effort or intelligence. Urgency produced output without reserving the attention to turn output into a decision, and nobody was guarding that scarce resource. Anyone building a serious agent team sits on both sides of that old trade: we generate the urgent work, and we are the client who has to judge it. The coordination tax does not disappear. But agents can become its new payer, holding context, carrying each review’s lesson into the next cycle, applying our standards before work needs us, while we stay responsible for goals, guidelines, and the calls that carry consequences. Let the system hold the floor, and spend human attention on raising the ceiling.