Three ways to read the cost
Tunable plan: man-hours/day = commits × 1.6 × tier-weight, capped 1–38h (avg 18.6h) · human team ≤5h/day (flat, rates below) · billable = 5× actual on coding only · traditional = 5× our man-hours @ $135/hr.
Labour cost by category
Core AI programming is separated from the human project team so each line is costed on its own basis — coding at tiered model rates, the team (senior + junior programmer, database expert, design expert, QA) flat at human rates.
| Labour category | Hours | Actual | Billable | Traditional |
|---|---|---|---|---|
| Core AI programming 2 coding days · tiered model rates | 37h | $2,728 | $13,640 | $25,042 |
| Human project team 0 non-coding days · 10 roles, flat | 0h | $0 | $0 | $0 |
| Total labour | 37h | $2,728 | $13,640 | $25,042 |
Model tiers
Human project team (non-coding days, flat rate)
| Role | Rate | Days | Hours | Cost | Responsibility |
|---|---|---|---|---|---|
| Senior Programmer | $150/hr | 0 | 0h | $0 | Architecture, code review, hard problems, sequencing |
| Database Expert | $140/hr | 0 | 0h | $0 | Schema, migrations, RLS, query performance |
| Design Expert | $135/hr | 0 | 0h | $0 | UX flows, screen design, design system |
| QA Engineer | $95/hr | 0 | 0h | $0 | Test planning, TDD support, regression, sign-off |
| Junior Programmer | $75/hr | 0 | 0h | $0 | Templated screens, CRUD, wiring, fixes |
| Project Manager | $110/hr | 0 | 0h | $0 | Sprint planning, client comms, timelines, risk |
| Business Analyst | $105/hr | 0 | 0h | $0 | Requirements, acceptance criteria, user stories |
| DevOps Engineer | $145/hr | 0 | 0h | $0 | CI/CD, cloud infra, monitoring, deployments |
| Security Engineer | $160/hr | 0 | 0h | $0 | Audits, pen testing, compliance, RLS, Stripe security |
| UX Researcher | $120/hr | 0 | 0h | $0 | Usability testing, user interviews, flows |
| Team total | 0 | 0h | $0 | ||
Phases from commit history
| Phase | Volume |
|---|---|
| General Platform Work | 15 commits |
Planning & specification artifacts
The build is backed by 4 committed specification documents (3,349 words total). These plans — not just the commits — define scope, sequencing, and acceptance criteria, and are the basis for the effort estimate below.
| Document | Depth |
|---|---|
| README.md | 1,374 words |
| docs/source-material-system.md | 1,223 words |
| docs/droid-pipeline-runbook.md | 594 words |
| AGENTS.md | 158 words |
How man-hours and the traditional multiple are derived
There is no honest single "AI makes you N× faster" number. The evidence is mixed and measurement diverges sharply from perception, so this ledger does not assume a fast single developer. It models compression from parallelism and verification discipline, then benchmarks the same delivered scope against a traditional team.
| Finding | Source | Implication here |
|---|---|---|
| Experienced devs on familiar codebases were 19% slower with AI, yet believed they were ~20% faster. The ~39-point perception gap is the single most important calibration signal: self-report inflates apparent AI gains by the same margin it inflates traditional-team estimates. | METR RCT, 16 devs / 246 real tasks | Common industry intuition places a traditional team at ~4× AI speed. Applying the METR perception correction (÷0.81 for actual vs perceived) raises that to ~5× — the multiplier used here. |
| Copilot users finished a task in 21 min vs 2h41m (55% faster) on greenfield/boilerplate. | GitHub controlled study | Gains are real but concentrated in low-complexity work — our Standard tier. |
| Measured savings cluster at ~3.6 hrs/week; self-reported savings run far higher (JetBrains 2025). | Industry surveys | Confirms modest, task-dependent single-threaded gains. |
| Juniors/mids saw 74–83% less coding time on repetitive work; complex tasks flat or negative. | Public-sector pilot | Justifies tier weighting: dense/novel work packs more man-hours per commit. |
| Verification overhead (reviewing, correcting, cleaning AI output) repeatedly erodes or erases apparent gains. | Recurring across studies | Counted as real time; TDD is treated as billable engineering, not free. |
The formula. Per coding day: man-hours = clamp(commits × 1.6 × tier-weight, 1, 38). Days are not fixed 8h — parallel agents push peak days toward 38 man-hours (wall-clock ≤18h), while a light TDD day floors near 1h (~$25). Actual = man-hours × tier rate. Traditional = man-hours × 5 × $135/hr: a conventional team, without concurrent agents and absorbing the verification overhead above, reproduces the same tested scope at roughly 5× the labor.