When Every AI Agent Is Waiting for Your Review

The agents were busy while you had lunch. Now there are five things waiting: a diff for the app, a plan for the game to approve, a question about the book's chapter order, a failing test the extension agent wants you to look at, and a PR description to check.

Everything is waiting for you. You're the only one who can't run in parallel.

Part 6 of Busy, Not Shipping — when the review queue becomes the bottleneck.

The short answer#

• With several agents running, you become the bottleneck: every finished task waits for your review before anything moves.
• Reviewing in random order, as notifications arrive, means constant switching and shallow reviews.
• Batch reviews into fixed slots, and review one project at a time, finishing each before the next.
• Make reviews smaller by giving agents smaller tasks with a clear definition of done.

The bottleneck moved to you#

Before agents, you were the slowest part because you did the typing. Now the typing is fast, and the slowest part is deciding: is this right, approve or not, which option. Every agent you add creates more decisions.

That's why a day with many agents can feel like busy all day with nothing shipped: you're reviewing constantly, but reviews done in a rush between notifications don't finish anything.

Review in slots, not on notification#

Reviewing each item the moment it arrives means constant switching and shallow attention. Instead:

• Fixed slots — say, late morning and late afternoon. Let the queue build in between.
• In each slot, go project by project. Finish all reviews for one project, then move to the next.
• Finish a review before starting the next: approve, reject with a reason, or ask one precise question.

Agents can wait an hour. A careless approval can cost you a day.

Make each review smaller#

The size of a review is decided when you give the task, not when you receive it. Smaller tasks make smaller reviews:

• One task, one outcome. "Add the tip presets screen", not "improve the app".
• A clear definition of done. "Tests pass and the screen matches the sketch."
• Small versions. The rhythm from ship small versions keeps every diff reviewable in minutes.

Know what's waiting without asking#

Part of the queue's weight is not knowing what's in it. Keep the waiting items visible the same way you keep status visible — one line per task, with "in review" as a phase. Then each review slot starts with a glance at the list, not a tour of every window, like the morning round in checking every repo by hand.

Run fewer agents than you think#

The practical limit on parallel agents isn't your machine or your plan. It's how many results you can review properly in a day. If the queue is always full, run fewer agents and review better.

Keep reading#

• Busy All Day With AI Agents, but Nothing Shipped? Here's Why
• Checking Every Repo by Hand Every Morning? There's a Faster Way
• Ship Small Versions: A Release Rhythm for AI-Built Apps