Building VialMath
We're building a real product in public — and showing the system behind it.
VialMath is a one-person software business run with a governed AI operating system — AI researchers, builders and separate reviewers. By rule, the agent that builds a change is not the one that checks it. This is how it actually gets built: receipts, constraints, failures and measurements.
Looking for the app? VialMath keeps your protocol coherent when one input changes — the product is over here. This page is the story of the build.
Why build in public
Transferable systems, with receipts.
Transferable, not proprietary.
We teach systems you can use on your own project — not private secrets, not a guru's "one weird trick".
Receipts over hype.
Every method comes with the constraint that shaped it and the measurement that tested it. Experiments are labelled experiments.
Shown, not claimed.
We show the method and the record of what it produced.
The operating system
Five things we run, and what each one is for.
Described at the level that transfers. The internal controls, prompts and infrastructure stay internal; the shape of the system is the lesson.
-
An AI operating system with distinct jobs
Research, building, review and governance are separate roles with separate briefs. The rule that holds everything up: a chat that authors a change does not approve it. Status comes from evidence, not from the author's report.
-
A small business built from a recurring problem
A recurring arithmetic problem became a utility. The decisions worth showing: free versus paid, one-time versus subscription, and starting narrow instead of building a platform.
-
Product discipline
Issue-based development; exact-successor builds; negative controls that prove a test can fail; regression suites; review by a second agent that did not build the change. "Tests passed" is not the same sentence as "release-ready".
-
A growth laboratory
Hook research, experiment ledgers, reads on fixed clocks, and the difference between views and intent. No winner is declared from weak data.
-
Operational transparency
What the lanes did, what failed and how it recovered, and what changed because the evidence contradicted an assumption.
What lives here
Two things, as they're published.
The log
Episodes drawn from real operating events — a build that failed a gate, an experiment that didn't pan out, a number that turned out to be ours — with the sensitive details removed and the lesson kept.
The toolkit
Free templates for people building small software products with AI coding agents, alone or in a small team, with every example invented: an agent task brief, a second-reader checklist, an experiment ledger, a hook analysis worksheet, a launch-readiness checklist.
The free templates are up. Until the first episodes are, the changelog is the public record of what shipped, build by build, and the narrated walkthroughs are on YouTube. No newsletter, no membership, no counts we can't stand behind.
Relationship to the product
VialMath is the live case study: this is the method it is built with.
VialMath itself stays what it is: a content-free, local-first protocol calculator and tracker that computes from the numbers you enter and never tells you what to do. Nothing on this page changes that, and nothing here is a promise of income or outcomes. We show what we tried, what broke, and what the evidence said.