Case study · The build system · Build economics

Move the assumptions yourself

The AI-native workflow case study puts a range on what building the app with Claude saved. Here's the calculator behind it: the measured facts are fixed, the only thing you're changing is the by-hand estimate. Drag the sliders and watch the savings recompute.

The measured facts are reconstructed from Claude Code's own session telemetry and committed to the repo as an aggregates-only snapshot. The dollar cost is self-reported and a floor. The PRs, lines of code, and velocity are git-verifiable, and reproducible from the repo's history.

Merged PRs155git-verifiable

Lines of code26,693git-verifiable

Supervised hours~110over 29 days

Claude cost$2,485list price · completed only · self-reported

Snapshot taken 2026-07-21 — frozen figures, not live ones.

By-hand hours per PR2h conservative

Engineer rate$100/hr

Claude cost basis

Time saved200 hrs≈ 5.0 weeks

Engineering time avoided$29kby hand: $31,000

Return on spend12.5×hand cost ÷ Claude cost

The by-hand side is an estimate — 155 merged PRs × 2h each, against the ~110 hours actually supervised. The Claude cost is self-reported from session logs and a floor, because the CI code-review agent's usage is not in those logs. Only the PRs, lines of code, and velocity are independently reproducible from the repo's git history.