AboutLehi, UT

One person, and the machinery that keeps him honest.

James Brady builds AI systems that show their work, from Lehi, Utah, across two entities: Utlyze, the studio, and New Reward, the agency. The page below says which entity delivers what, and answers the fair question about reliability without jargon.

The short version

James Brady builds AI systems that show their work. One person, operating at fleet scale, documenting what actually works. Based in Lehi, Utah.

The personal history behind this work is not published here yet. Neither is the multi-year goal, in James Brady's own words. A photograph of James Brady has not been published here yet either.

Who does what

Three names show up across this site, and they are not the same thing.

James Brady is the person. The theories, the open-source projects, and the writing are mine.

Utlyze is the studio. It builds products and takes on custom software and systems work.

New Reward is the agency. It does the visibility work: measuring how findable a business is in search and inside AI assistants, and then fixing it.

I operate both. When a page on this site says a client engagement, it names the industry and not the company, unless that company has given written permission to be named.

How these two names sit together as legal entities is not confirmed on this page yet, so the split above is the working description rather than the paperwork.

Reliability

How one person checks work before it ships.

This is the fair question, answered in plain words rather than in process vocabulary.

If most of the building is done by AI agents running in parallel, what stops bad work from reaching a client?

A claim is not proof. An agent saying it finished is a report, not a result, so the finished thing gets checked at the place it actually lives, not in the chat that produced it. If a change is supposed to be on the site, the site is what gets read. If a report is supposed to have gone out, the sent folder is what gets read.

What does “done” mean here?

Work moves along a ladder, and each rung is a different claim: open, then checks passing, then reviewed, then merged, then deployed, then checked live, then approved for a client to see. Saying a higher rung than the true one is treated as a defect, not a rounding error.

Who grades the work?

Whatever grades the work is separate from what produced it, and the standard is frozen before the work starts. If the thing being tested can edit the test, the test decides nothing. The frozen set includes the test files and the scripts the check depends on, and a change to any of them is compared afterward.

What happens when something cannot be measured?

The answer is “unknown”, and unknown is written down as unknown. It never quietly becomes a zero, an estimate, or a confident sentence. Missing is not zero.

Machines hold the gates, not memory. Required checks run on every change. A merge queue tests changes together, in order, before they land, so two changes that each pass alone but break together get caught before a person sees them. None of this depends on anyone remembering to run something.

Numbers are computed, not typed. Every number on this site is derived from its source when the page is built. The old version of this site displayed a hand-typed count that was wrong by a factor of eight. That class of mistake is now impossible here, by construction.

Irreversible things stop for a human. Sending a client message, spending money, publishing, deploying, deleting data. Agents prepare those. A person approves them. That gate is the design, and not something I am working around.

The throughput figure has been measured but is not published here yet, so this section stays qualitative.

What I use, and what is not mine

OpenClaw is a third-party project. I run and extend a self-hosted instance of it, with custom skills, a gateway, and hardware hookups. I did not build OpenClaw.

architect-loop is a fork of a project by Dan McInerney. I run a heavily customized copy daily and have contributed a fix upstream. The original design is his.

I mention both because a portfolio that quietly absorbs other people's work is exactly the kind of thing this site is supposed to be the opposite of.