What this would have cost without AI

One operator built Wolfberg and everything under it: a multi-tenant runtime, a six-agent modernization pipeline, a second deployable business, the marketing site, a 1,500-page operating brain, a five-day curriculum, and the company itself. It took 237.36 logged hours and under $50 of metered compute. Below is the like-for-like cost of a team doing the same work without AI. Every figure on our side is computed from instrumented data. Every figure on the team side is built from published industry rates. Sources are inline and listed at the end.

Companion receipt: this page is the cost receipt. The coverage receipt (where we stand on all 43 named AI problems in the industry, with status and the system doing the work) is at /the-43.

The comparison, three ways

The same total scope, delivered by a no-AI team, under a skeptical / central / generous reading. The multiplier is the team's man-hours divided by the operator's 237.36 verified hours.

The one line: between roughly $1.5M and $7.1M of equivalent build, work that would take a team ~15,000 to ~40,000 man-hours, was produced by one person in 237.36 hours for under $50 of metered compute. Even the most conservative reading is a labor compression of about ~65×.

What got built

Everything a team would have to reproduce. Counts are from the live metrics job (see "Our receipts" below).

• Keystone — multi-tenant AI-workforce runtime. 6 of 6 production modules, 108 AWS resources, p95 317 ms at 173 RPS.
• Refactory — six-agent legacy-modernization pipeline. Proof pilot shipped and verified.
• The marketing site — built and load-tested, with an embedded live assistant and a visualization of the operating brain.
• A second deployable business — a complete, tenant-ready property-management platform the model produced as a side output. Built and shelved; not operated (0 paying tenants). Counted at build cost only.
• The operating brain — 1,656 structured pages and 162 typed memory files of operating knowledge, plus the gating and coordination systems that run it.
• The curriculum — a five-day AI-operations program with full materials.
• Sales and positioning — seven decks across many revisions, the positioning briefs, market research, the diagram library.
• The company — formation, EIN, banking, payments, SMS, tax and back-office stand-up, partner agreements.

Verified output: 104,154 lines of code · 17,897 lines of documentation · 1,656 structured pages · 940 commits · 162 memory files · 1 live production deploy. Metrics job run 2026-06-18.

Our receipts — how we measured

None of our numbers are self-reported. Each is produced by an automated job from instrumented data and published, unmodified, to this site.

The team's receipts — how we estimated

Bottom-up by output. We take each deliverable's verified size and divide by a published industry productivity rate, then cost it two ways. This is the average case; the conservative and liberal cases move the productivity and coordination levers (see methods). Productivity rates favor the team where there is a choice, so the estimate runs conservative.

The three cases, side by side

Costing: loaded = US-market base salary × 1.33 overhead (MIT Sloan / Hadzima rule for fully-burdened headcount). Consulting = sourced contractor and fractional rates for each role (senior architect $200–500/hr, senior engineer $100–175/hr, technical writer $30–75/hr, designer $50–150/hr, PM $100–150/hr). Calendar accounts for the mythical-man-month effect (Brooks's Law): you cannot divide the hours by headcount and finish proportionally, because communication overhead grows with team size and architecture must precede implementation.

The honest accounting

Three ways to state the gap. We separate hard cash from the value of the operator's own time, because conflating them is how this kind of number gets dishonest.

Read the third row, not the second. The cash-only ratio is real but unfair: it ignores that one person spent 237.36 hours. Pay that operator fully at a senior market rate and the all-in is still roughly $60–100K against a team's $1.4–3.8M — a ~20–45× compression after honest accounting. That is the number that survives scrutiny.

Methods & honesty

1. Hard cash is kept separate from imputed time. Only $49.84 left a bank account. The value of the operator's 237.36 hours (~$60–100K at market rates) is imputed, not spent, and is labeled as such everywhere above.
2. The second business is counted at build cost, not as a going concern. The property-management platform is built and tenant-ready but shelved, with zero paying tenants. No revenue or operating traction is claimed.
3. The three cases move two clearly-stated levers. Conservative assumes a fast, lean team (~50 net LOC/dev-day, minimal coordination) and a generous haircut on the code that counts as real effort. Liberal assumes a slower pace under full enterprise process (~28 LOC/day) and heavier coordination. The truth is a range, and the range is shown.
4. The estimate is most sensitive to two inputs, and both are disclosed: the lines-of-code-per-day rate (a swing from 25 to 50 moves engineering hours by about half) and the share of the codebase that is hand-authored application logic versus infrastructure-as-code and configuration. The conservative case already discounts for this.
5. This is a self-assessment, not a certification. Our security posture follows a documented NIST 800-171 self-assessment with a strong implemented control base (account-wide logging, threat detection, MFA, encryption at rest, strict transport and content-security headers). It is not third-party audited, and we do not claim "compliant."
6. It is an estimate. A counterfactual team's effort cannot be measured, only modeled. We have shown the model, sized every input from a published source, and run it conservatively. Swap in your own assumptions and the floor still lands one to two orders of magnitude up.

The coverage companion: beyond the cost receipt, our status on every named AI problem in the industry, sourced and honestly hedged, is at /the-43.