case-studies
Case Studies
Philosophy pages say "should". Case studies say "did", with numbers and receipts.
Three of them: the rig, the SaaS, and the engineer who learned the workflow and sold work with it.
The three studies
| Study | Subject | Headline |
|---|---|---|
| One week inside a governed agentic coding rig | Turbo Rig — my own engineering workflow | 180 PRs, 967 cross-model gate reviews, 6.57B tokens, one audited week, zero agent merges |
| Building and operating an AI SaaS in production | Ancuria — real-estate SaaS for Mexico | Conekta payments, CFDI, an AI assistant, and the operational discipline that keeps it up |
| From first container to sold work | Jan Lafko — independent developer, 22 | Gore-Tex company website delivered with the workflow; 220 quality gates donated back |
How numbers are handled here
Every metric in these studies is traceable to its writer — gate logs, deployment records, or the published weekly stats. Where a number is a measurement of my own workflow, it is labeled with its measurement window. Where I have no number, the sentence is qualitative or absent. That policy is itself part of the point: AgentOps without measurement is just opinion.
Talk it through with someone who runs this stack on his own systems every day.