The Org Chart Nobody Hired Everyone is arguing about whether AI takes jobs. Almost nobody is measuring what one person plus AI actually produces. So we measured it. Same method as the energy piece we published at indelible.one/energy: count the real artifacts, label what is measured and what is estimated, publish the receipts. One thing before any numbers. Nobody in this story was displaced. There was never a team to replace. The builder is a solo founder who runs a pool company; the counterfactual to this system was not a hired engineering org, it was this not existing. Hold that frame. It matters more than the dollar figures. What We Counted (measured) Five months, February 18 to July 15, 2026. One founder, zero employees, an AI stack. From the artifacts on disk and on-chain: 102,294 lines of production code across 355 files, with generated bundles, backups, and vendored code scrubbed out before counting. Another 15,923 lines of tests across 100 files, and 58,799 lines of technical documentation across 310 markdown files. About 177,000 authored lines in total. 55 npm releases since February, ten of them in the first fifteen days of July. 117 completed goals in a dated lifecycle store, running at 10.4 per week in the current era. And this is the part a line count misses: the system is not just built, it is operated. A 7-node server fleet plus a home bridge, with six staged fleet-wide production rolls in two and a half weeks, each one canary-first with per-box rollback. Two self-run blockchain validators and two chain indexers, with failover proven live. Four automated alert paths. An automated review swarm of 26 agents whose credential gate has issued 14,329 lifetime verdicts, rejecting 5,733 saves before a secret could be written somewhere permanent. Behind all of it, 8,587 on-chain records covering 5,289 working sessions and 442,320 messages. Those are the measured facts. Now the estimates, labeled as such. The Roster (estimated) We asked a blunt question: which distinct engineering functions are demonstrably being performed here, and what is the smallest honest org chart that owns each one? Not lines-per-engineer, which is a discredited heuristic. Function by function, evidence first. The study's answer: 9.25 full-time equivalents on the conservative read, 13.5 at fair value. Backend and protocol engineering (a 29,121-line MCP core with cross-runtime cryptography). Blockchain infrastructure (a 25,194-line bridge protocol, a sovereign indexer with a streaming binary block parser, two validators in production). SRE work (the fleet rolls, the monitoring). Frontend (29,632 lines of React including an in-browser WASM embedding worker). A web server. An ML stack (local semantic search with browser-to-CLI parity proven to seven decimal places). QA, security, technical writing, product management, release management, support. Every row in the study cites the artifacts that prove the function is real. One structural note. Real companies cannot hire half a security engineer, and an on-call rotation for a 7-node fleet needs at least two humans. A conventional org attempting this scope at this cadence would plausibly post 12 to 15 job openings. What That Payroll Costs (estimated, sourced) Now convert the roster to dollars. Two compositions, per the study: conservative (9.25 FTE, priced at national medians) and fair value (13.5 FTE, priced at market rates for the specialties involved). The conservative column uses BLS Occupational Employment and Wage Statistics medians, May 2025 reference period: software developers $135,980, QA analysts $104,300, information security analysts $129,180, technical writers $90,390. Roles BLS does not break out (SRE, ML, frontend) are priced at the plain software developer median, which underprices them and keeps the number honest. Blockchain engineers, which BLS does not track at all, get the Glassdoor and ZipRecruiter convergence of about $146,000. Product management uses Glassdoor's $150,990. Multiply each role's salary by its FTE fraction and sum: $1.21 million in base salaries. The fair-value column prices the engineering specialties at market and keeps the BLS medians where market data was thin, which understates it: the levels.fyi 2025 report's mid-level software engineer median of $226,000, $205,000 for SREs, $272,500 for ML engineers, $230,000 for product managers. One label matters here: levels.fyi figures are total compensation, salary plus stock plus bonus, not base salary, and the report skews toward large tech companies. Sum: $2.51 million in total compensation. Salaries are not what employees cost. Per the BLS Employer Costs for Employee Compensation data, wages are only 69.9 percent of what a private employer actually pays; benefits and legally required costs are the other 30.1 percent. That implies a 1.43x fully-loaded multiplier on base wages, so the conservative column gets 1.3x to 1.43x. The fair-value column is total compensation that already contains bonus and stock, which ECEC counts on the benefits side, so applying the full multiplier there would double-count; that column gets a reduced 1.2x to 1.3x. Both still exclude recruiting, equipment, and office overhead. The arithmetic: Conservative: $1.21M x 1.3 to 1.43 = $1.57M to $1.73M per year Fair value: $2.51M x 1.2 to 1.3 = $3.01M to $3.26M per year Prorate to the five-month window (actual span 4.8 months; we round up to 5/12, which slightly overstates these figures): Conservative: $656,000 to $722,000 of labor consumed Fair value: $1.25M to $1.36M of labor consumed Said plainly: the output of the last five months, priced as labor, runs somewhere between $656,000 and $1.36 million. The input was one person's time plus AI subscriptions. For this build, the price of a small engineering org's worth of output fell to consumer-subscription money. The Caveats, Stated Plainly Because that is still the whole point. Bus factor is 1. One brain holds the map. A real team has redundancy this operation does not. Test coverage is uneven. The bridge carries 9,680 lines of tests; the 29,632-line frontend has 70 lines of tests sitting next to it. Roughly ten ships went through adversarial review gates, but review by AI agents the operation built itself is not an independent audit. Several subsystems have never seen outside eyes, and the census flags its own open security item. Some of those 9 to 13 FTEs are coordination overhead. A real team pays a heavy tax in meetings, interface negotiation, and integration bugs at team boundaries. Nobody here performs that work, which is partly why the output was possible. The honest reading: the output matches the team; the structure does not. AI wrote most of the lines. We think that is the finding, not the gotcha. The census shows the human performing the roles organizations pay most dearly for: 117 goals scoped and accepted, every publish and deploy behind an explicit human gate, product pivots and vetoes logged and dated, 54 written handoffs covering about 70 percent of calendar days since May 2. This is a managed process. The founder is the engineering manager, product lead, and final QA gate of a team whose engineers happen to be models. And once more: nobody was displaced. These payroll figures are what the labor would have cost, not what anyone saved on salaries, because those salaries were never going to be paid. Capability expanded. No job contracted. The Point The energy article measured what AI forgetting costs. This one measures what AI remembering builds: a full protocol, fleet, and product surface, held together across five months by a system that does not lose its context between sessions. Indelible is that memory layer, built on itself, and the build is the demonstration. The full study, with every census line, the roster evidence, and every salary source with links, is at indelible.one/team. The energy receipts are at indelible.one/energy. Check our math.