Where developer time goes

Quick reference

Waiting for CI

• A ten-minute run four times a day is forty minutes
• Queue time in front of a runner is not build time
• Track the median and p90 duration per job
• Cache dependencies and split the slowest job

Review latency

• The wait for a first reviewer, not the approval
• A large diff gets postponed, then skimmed
• Keep a review rota and a first-response target
• Measure time to first review, not time to merge

Environment setup

• A new laptop to a running stack can take days
• Every manual step drifts from the documented one
• One command should build, seed and run the stack
• Containers make setup repeatable, not instant

Context switching

• Every interruption costs a rebuilt mental model
• An interrupted review restarts from the first line
• Protect two-hour blocks and batch the pings
• Fewer open pull requests means fewer switches

Explained

Waiting for CI

Waiting for CI is the most visible queue. A ten-minute pipeline run four times a day is forty minutes at the desk, and queue time in front of a runner is not build time. Track the median and the ninetieth percentile per job, cache dependencies, and split the slow job that blocks everything else.

Review latency

Review latency is the wait for a first reviewer, not the time to approve. A large diff is postponed until Friday and then skimmed, which is how a two-day wait becomes a shallow review. Keep a review rota, set a first-response target people can meet, and track time to first review.

Environment setup

Environment setup is the cost nobody budgets. A new laptop to a running service can take days of undocumented steps, and every manual step drifts. One command should build, seed and run the stack; containers make that repeatable, not instant.

Context switching

Context switching is the tax on all three. Every interruption costs a rebuild of the mental model, and an interrupted review restarts from the first line. Protect two-hour blocks, batch the pings, and keep fewer pull requests open per person.

In one line

Measure the wait

$gh pr list --json number,createdAt,reviews