11. StackNative as the case study

Draft. This is the chapter that cannot be faked, and the reason to keep it honest is that it is the only part of this book a sceptical reader will fully trust.

StackNative is a company whose CEO is an AI. One human, Marty Chang, sits on the board and handles approvals, incorporation, and anything requiring a legal person. Everything else is run by an agent working in five-hour shifts, with durable memory in files, a Discord channel to the board, and a task board.

What this chapter will record, with real numbers and dates:

• The operating model. Shifts, handoffs between sessions, memory as the substitute for continuity, and what breaks when a session ends mid-task.
• The economics. Actual cost per reading and per consultation, actual monthly spend against the ceilings, and where the estimates were wrong.
• The checkpoints. Which decisions genuinely required the human, which ones we thought would and did not, and the one standing rule: act on identified gaps, escalate only unbudgeted spend.
• What went wrong. The DNS outage caused by changing nameservers with a stale delegation signing record. Restarting the service from inside the session that was doing the restarting. Publishing prices that had never been tested against a customer.
• What is still unproven. As of version 0.1: no revenue, no delivered assessment, no pilot. The method in Part II is a hypothesis derived from experience and reasoning, not yet from a book of cases. It will be revised from the first ten real assessments, and the revisions will be visible in this repository’s history.