A friction log is a running record of interruptions and unnecessary work in a developer’s workflow. Write down waits, context switches, manual steps, and moments of confusion as they happen. The point is not to count how busy you were; it is to find recurring obstacles worth removing. section 10.5.1 section 10.6
Turn the record into one small improvement:
- Log without filtering. Include what happened and, where possible, roughly how long it cost. section 10.6
- Review the entries for repeated or costly friction. Group items such as waiting, rework, searching, context switching, and manual work; choose one you can change. section 10.5.1 section 10.6
- Record a baseline, make the change, and keep logging. Measure both the friction it removes and any new friction it creates before deciding whether to keep it. section 10.6
A CI wait that starts locally
Suppose lint errors repeatedly reach CI and trigger 18-minute reruns. A log makes the pattern visible. Add a pre-commit lint hook, then record errors it catches and any delay the hook adds. In the worked cycle, lint-related CI failures fell from two per week to zero; the hook took eight seconds on a large commit. The decision to keep it rested on that comparison, not on the assumption that every automation helps. section 10.6
Listing 10.3 Friction log
# My Friction Log – Day 1
08:45 Opened IDE. Waited 3 min for indexing.
09:10 Ran tests. 2 tests failed — both flaky.
Re-ran. Passed. Wasted 4 min.
09:30 Started working on PROJ-456. Realized I
needed the latest schema. Searched docs
for 8 min. Found an outdated version.
Asked teammate on Slack. Waited 20 min
for reply.
10:40 Ran integration tests locally. Had to`
`manually start three services first.`
`Forgot the config flag for service C.`
`10 min of setup a script would prevent.`
11:15 CI failed. Read the error. It was a linting
issue I would have caught locally if I had
a pre-commit hook. Pushed a fix. Waited
18 min for CI to re-run.
14:00 Context-switched between 3 PRs during
review. Lost track of what each one was
trying to accomplish. Re-read code I'd
already read. 15 min wasted.A time-stamped day of IDE waits, flaky tests, outdated documentation, integration-test setup, and CI reruns shows the level of detail needed to spot patterns. section 10.6
The same habit applies to AI-assisted work: log unexpected output, time spent adjusting generated code, and workflows disrupted by model upgrades. Review the pattern before changing prompts or adding specifications. A friction log is a sensor for improvement, not an improvement by itself. section 14.5
