The Lead

Most AI-coding stories are about one output: a model wrote a function, an app, a demo. This one is about the opposite. The interesting artifact here isn't any single program — it's the repeatable method that produced all of them, and the fact that the work got faster as it got more complex, not slower.

The maker is Shea Gunther. Forty-eight, then a framing-and-renovation carpenter in New Gloucester, Maine, working from a one-room RV on a $200 laptop. Zero professional software experience. He started in March 2026 by typing something close to “Hey, you can code, right?” into Claude — called out of the trade to do it, and has worked on Loop MMT — Multi-Module Theory — nearly every hour since.

What came out over the following months isn't a pitch deck. It's shipped, running code you can read: an eighteen-thousand-line browser computer in a single file with no dependencies; a real deer-processing business's order system, spec to shipped in four days; a personal-data platform you host yourself; games; a solver. Eighty-seven single-file tools by that May, each under three thousand lines, each following the same five rules. The apps are up now, source included.

The differentiator worth writing about: the method solves the thing that makes AI coding reset to zero every session. Loop MMT keeps its own memory in a git repo, so a fresh model each session reconstructs the full working context and builds on top of the last session instead of starting over. That's why the work compounds — and it's the part a “the AI wrote it” framing misses entirely.

It is honest about its ceiling. Nothing here claims the machine is doing the thinking, and nothing here is a finished company — the site says pre-launch because it is pre-launch. The argument for why any of it matters — personal data sovereignty — is one door over, and the whole build in order is another.

The claim isn't “AI can code.” It's “here is a reproducible way to build real software with it, from a laptop, and here is the pile of working software that came out.”

The facts, checkable

Every figure below is derived from the project’s own git record and published on the site, not asserted here. The links go to the primary source so you can verify each one independently.

Who

Shea Gunther — a framing-and-renovation carpenter in New Gloucester, Maine, with no professional software background before this. Full background: Who.

What it is

Loop MMT (Multi-Module Theory) — a method for AI-assisted work that keeps its long-term memory in a git repository, so a fresh, stateless model reconstructs the full working context each session and builds on the last one instead of starting over. The one-page version: The Thesis.

When

Began in March 2026 with a single message to Claude. The dated build, in order: Timeline.

What shipped

Running, readable, single-file code — among them an eighteen-thousand-line browser computer with zero dependencies, a working deer-processing order system built spec-to-shipped in four days, a self-hosted personal-data platform, games, and a solver. Source is included on each. Browse them: The Apps.

Scale of the record

The site reports a byte-derived work record — on the order of tens of thousands of commits across thousands of sessions — computed from the git history itself, not hand-counted. The exact current figures, and how they’re derived, are on the Receipts.

The method, in five rules

The constraints every tool follows (single file, no dependencies, and the rest): The Five Rules.

Verify it yourself

This page is a starting point, not the story. The story is verifiable at the source, and the project is built so you can check it without taking anyone’s word:

• The Receipts — artifacts the system produces on its own (a timesheet derived from the git log, a session integrity badge, a close pegboard), lightly cleaned and byte-derived. Read the receipts →
• The full corpus — the Creed, the Protocols in Force, the glossary, and every governing document, readable in full. Open the corpus →
• The source is canonical — on every page of this site, the rendered view defers to the source file; when the two disagree, the source wins. Nothing here is a mock-up of work that doesn’t exist.
• How the site itself was built — this website is a specimen of the method, built by it. How This Site Is Made →
• The argument for why it matters — personal-data sovereignty, stated plainly. The Case →

What this is not — on the record

An accurate story needs the caveats as much as the claims, so here they are, plainly and quotably:

“Nothing here claims the machine is doing the thinking, and nothing here is a finished company. The site says pre-launch because it is pre-launch. What’s being shown is one method for building with AI, working — not a product, not a company, and not a claim that the AI did it alone.”

If a detail here doesn’t check out against the source, that’s a bug worth reporting — the whole point is that it should.

Reach the maker directly at shea@loopmmt.com.