Coordinate and act
The Cap
Cap how much work can be in flight at the stage where the queue forms, with an agreed rule for what happens at the cap.
Borrowed from work-in-progress limits from lean manufacturing and kanban. Read the original source.
Changes a rule or role · An hour or more to set up · Anyone on the team can start it · Start here for this problem
Try this first
Count what is open at the review stage. Set a number. Agree that nothing new starts until something clears.
Use it when
Output volume is up, review is behind, and the response so far has been to ask the reviewers to go faster.
Skip it when
It needs a workflow with identifiable items moving through stages. On genuinely continuous or one-off work there is nothing to count.
How to introduce it
Count what is actually open right now at each stage. Not what is assigned, what is in flight. Wherever the pile is deepest is your constraint. We are going to put a number on that stage and decide what happens when we hit it.
How to show up
Get the real count before anyone argues about the number. Push back on the first cap being generous, and then push harder on the stop rule, because a cap with no agreed consequence is a suggestion.
How long it takes
Two hours to count and set. Half a day if the room needs to argue about where the constraint is.
What makes it hard
The cap feels like it will slow the team down, and for the first week it does, which is when people abandon it. Someone senior will also ask for an exception on day three, and how you handle that decides whether the cap survives.
What it looks like when it's working
Items finish faster than they did before, and the team stops starting things it cannot get to. Rising counts with the cap still nominally in place means it is being ignored.
How long until it sticks
Three or four weeks before the team stops feeling the constraint as a restriction.
How you know it stuck
The cap gets reviewed against actual throughput rather than defended or ignored.
The idea behind it
Starting more work makes everything slower. The instinct when things are late is to start earlier, and that spreads people thinner and stretches every item.
Where it comes from
Lean operations. AI raises how fast work gets produced without raising how fast it gets checked. The checking stage becomes the constraint, and an uncapped constraint fills until quality drops silently. The team counts what is genuinely in flight, finds the stage where the queue forms, sets a number, and agrees what happens when the number is hit.
Pairs with
The Backpressure Signal, which handles the case the cap misses. The pass, which names who owns the final check.
Evidence
Working with WIP limits for kanban, Atlassian
Maybe we shouldn't be reviewing all this code, martinfowler.com Thoughtworks' chief technology officer reports that the size of code changes has grown sharply with AI. She argues that reviewing everything line by line no longer works, and that teams should build quality checks earlier and keep human review for high-risk changes.