Research paper · publishedResearch paper
How do you verify a stack that claims to build and run a company? Testimonials are selected. Benchmarks are gamed. This paper argues for the strongest available test: the company that sells the stack runs on the stack, in public, with receipts anyone can inspect. We describe the arrangement at Armalo, where the company is built, marketed, sold, and supported on its own product; we state precisely what that proves and what it cannot prove; and we give the disciplines that keep self-operation honest, including independent checks, external ground truths like the bank balance, and proof-triggered retirement of every temporary substitute. Self-operation is not a stunt. It is the verification strategy the category demands.
The strongest available test of a stack that claims to build and run a company: the company that sells the stack runs on the stack, in public, with receipts anyone can inspect.
Open the note →Research paper · publishedResearch paper
AI agents now do real work, and three gaps keep that work from being trusted: evaluations that pass while customers get hurt, evidence that ranks last among governance activities, and approvals that happen after the fact or not at all. This paper defines the smallest mechanism that closes all three gaps at once. Every job has a boundary, stated before the work starts. Every action leaves a trail, written while the work happens. Every outcome gets checked, against the real world rather than the agent's own report. We call the composition of the three a receipt, and we argue the receipt, not the score, is the unit a business runs on. This is the mechanism layer of the Armalo Layer: the trust layer for AI that does real work.
Every job has a boundary, stated before the work starts. Every action leaves a trail, written while the work happens. Every outcome gets checked, against the real world rather than the agent's own report. Compose the three and the result is a receipt, and the receipt, not the score, is the unit a business runs on.
Open the note →Field study · activeResearch brief
A field study of the boundaries enterprises need before agents can safely touch production work.
The enterprise adoption question is not whether agents can code. It is whether the factory can govern their access and preserve the reason for every change.
Open the note →Product experiment · runningExperiment note
Testing whether security and compliance evidence can be produced by normal work instead of a last-minute reporting sprint.
The best evidence is the evidence the system needed in order to operate safely in the first place.
Open the note →Build log · publishedBuild log
Notes from designing a software factory for many businesses, where one team's speed cannot become another team's risk.
Keeping businesses separate is not a database feature. It is the rule that every identity, action, and artifact belongs to the right customer.
Open the note →