The Agent Standup: Practice Note
In brief
What the agent standup is, what it isn't, and how to run it every day.
A one-page companion to The Agent Standup template and section 10 of The Standup Playbook.
Every agent I run keeps working whether I'm watching or not. The Agent Standup is how I check in on them. An agent here means any AI worker running a task on its own: a coding agent, a drafting worker, or a loop that dispatches other agents and keeps going without me.
The Agent Standup is a daily report, generated instead of spoken, built on the same practice as a team standup: what happened, what's running, what's stuck, what needs a decision, what's planned. The participants are agents; the questions are the ones a team answers every morning. Nothing stops for the page. The loops that ran yesterday are still running today, and I read it the way a lead walks a board while the team keeps working.
I run it once a day. On a normal day I read the one-line summary, check what needs a decision, and scan the rest. On a quiet day most sections shrink to a single line: “none,” “nothing stalled.” The rule that doesn't bend is that the page runs every day regardless. A skipped day breaks the one thing a daily cadence buys, which is seeing today against yesterday against the day before. Skip it once because nothing happened, and you stop noticing when something finally does.
The section asking for a decision needs the most restraint. Start with a budget of three decisions a morning. A page that lists everything needing attention, with no limit, becomes another queue. Items past the budget fall to tomorrow. Track how many decisions actually show up over a week, and let that count set the number; three is a starting point, and yours may be different.
Every line on the page is marked claimed or verified. Verified means something other than the agent checked it: a test passed, a commit exists, an exit code came back clean. Claimed means the agent said so about itself, including the confidence number every worker already reports on its own runs. That confidence line is worth reading, and it is never sufficient on its own. The studies behind this design found that agents' reports on their own work overstate success, so the page keeps claimed and verified apart, and a self-report gets a line on the page without getting the decision.
The page borrows its logic from the Standup Playbook's board-walk (sections 2 and 6), and the playbook's section 10 is where this page is introduced. The sections here read in the opposite order from a live standup: a live standup walks toward what's closest to done, while this page opens with what needs a decision and what's gone quiet, because nobody is in the room to raise it as it happens. What carries over is the per-item question, what would stop this from being done by tomorrow, and the habit of aging stalled work instead of letting it sit unremarked.
For now every line is filled in by hand; a generator that fills the machine-fillable lines from the loop's own state files is in progress. The page runs every day either way, short on quiet days and longer on bad ones.