Where bottlenecks move
A team may spend less time writing a first patch but more time checking whether it fits the architecture. Another may get fast answers but wait for requirements, test data, or a reviewer who understands the change.
Bottlenecks can appear before a session, inside an exploration loop, or after an agent reports completion.
Use the trajectory to locate the constraint
Compare where context arrives, where retries repeat, when verification begins, and where work is handed off. Keep task complexity and system familiarity in the comparison.
This creates a more useful diagnosis than saying a team is slow or a tool is inefficient.
Change one part of the system
The response could be clearer task framing, a better test harness, earlier review, stronger ownership, a model change, or a rule for adding evidence after failure. Choose the intervention that matches the observed constraint.
Then review comparable work again. A bottleneck analysis is useful when it helps test a change, not when it becomes another dashboard.
Frequently asked questions
Are bottlenecks always caused by AI?
No. AI may expose or move an existing constraint. Requirements, architecture, review capacity, and system complexity still matter.
Can a bottleneck be useful?
Yes. A constraint can show where an investment or workflow change will have the clearest effect.
TraceYield
Start with one real work trajectory.
Discuss the question you want to investigate with TraceYield and the context required to answer it.
Request a pilot