Catch-up — Measure the Workflow Before Replacing the Tool
A dramatic architecture is useful evidence of a constraint, not a template for everyone else.
- Willison reports that Anthropic’s main-branch history is almost 100 GB and a compressed working copy exceeds 25 GB.
- He reports roughly 25 million daily code checkouts for CI before developer environments, agent sandboxes, and research jobs are counted.
Catch-up retrospective covering sources published 17 September–13 October 2026. This is not a backdated daily edition; it closes the gap with an architecture account from the most recent reviewed source window.
Large teams make unusual tradeoffs because their bottleneck is unusual. Copying the solution without matching the constraint can buy migration risk, training cost, and lost capabilities without fixing anything important.
Choose one tool your team calls too slow. Record the failing operation, workload size, frequency, wait time, and workaround. Set a threshold that would justify replacement, then compare a smaller configuration or process fix first.
Put it to work 25 minutes
A measurable replacement trigger for one engineering tool
Write a one-page replacement trigger for one strained tool: current workload, observed failure, acceptable threshold, cheaper fix, and the evidence that would justify migration.
See a worked example
Use measurements already available from a development workflow. If a value is unknown, label it unknown rather than estimating it.
- Observed constraint
- Repository checkout / measured size / median and worst-case wait.
- Replacement threshold
- Replace when checkout time breaches the agreed CI budget for four consecutive weeks.
- Cheaper change first
- Test sparse checkout on two jobs and compare duration and maintenance burden.
Your turn
Name the operation and measured failure.
State what would make migration worth its cost.
Test one configuration or process change.
Your draft stays in this browser. No account needed.
Do not wait for catastrophic scale when a tool creates reliability or safety risk. Some constraints justify an earlier move even at modest volume.
Willison provides attributable reporting and links the underlying account; your own replacement decision still requires measurements from your workflow.
Did the cheaper change restore the workflow below its agreed threshold?
Watch: Observed wait time for the failing operation · Workarounds or incidents caused by the current toolKeep Going