
$40T National Debt. 1M Agents for the Principal, Humans for the Interest.
Split the debt into two jobs. One million AI agents pay the principal on a public schedule. Humans carry the interest. Then do the math honestly.
Engineering notes from ML Systems, a construction technology company in Rhode Island. Claims-based data models, multi-agent architecture, on-device vision, and what breaks when software meets a job site. Written by Salman Parvez, who runs the construction company and wrote the software. Every claim here is labelled MEASURED, MODELED or ASPIRATIONAL.

Split the debt into two jobs. One million AI agents pay the principal on a public schedule. Humans carry the interest. Then do the math honestly.
The model is rented. The window is authored.

Reviewing AI-written code is a bounded task — if the source exists. Naming the role that writes the source.

"Evaluators embedded inside the lab, with employee-level access." That is step one of Dario Amodei's We Must Pace the Frontier, published September 12. Most of the commentary treated it as a policy qu

On ML Systems' Master Ledger a fact is not a value. It is a claim — a value, plus who made it, plus the grade of evidence behind it — and several claims sit on the same entry. The ledger decides which

Four claims about one roof, and what the Master Ledger would have to do when a section is turned over and the layers start disagreeing with the file

Domain-scoped authority, reconciliation states, multiverification, and signatures that lapse when the content changes

A response to Marina Wyss's How I'd Learn Machine Learning in 2026, from someone who arrived at the same checkpoints from the construction side
Every table I have ever written starts from the same quiet assumption: that there is one right answer and my job is to store it. bedrooms: 3. Done. That assumption survives right up until two sources