00 a one-operator AI studio
One operator.
One engine.
Agents that don't make things up.
AJ AI is the system. Agent AJ is AJ — cloned into an AI operator that runs it. One machine, one loop, real tools: it builds and audits its own software, grounded in real data, with a human on every irreversible call.
01 what this is
The system, and the clone that runs it.
Two names, one idea. It ships real software because the operator never sleeps — and never gets to skip the audit.
The system
AJ AI
An engine on one machine that runs a simple loop — classify → route → execute → persist — over real tools: Claude as the kernel, a vault as canonical memory, executors that write actual code, and a scheduler that works nights.
engine · vault · router · audit pipeline
The operator
Agent AJ
AJ, cloned into an AI operator. It carries his context, his projects and his judgment — does the building, runs the audits, drafts the calls — and leaves anything irreversible for the human AJ to sign.
the clone builds · the human signs
02 how it works
One loop, running right now.
Every request — a build, a brief, an audit — takes the same path through the engine: a deterministic router and four honest steps. Here it is, live.
classify
Read the request. Decide what kind of work it is.
route
Deterministic router picks the model + tool — zero added LLM cost.
execute
Real executors do the work — code, research, audits.
persist
Everything written down — vault is memory, ledger is audit.
03 how it's built
Nothing irreversible ships without passing four gates.
Trust isn't a vibe here — it's a pipeline. Code, deploys, anything touching money: every one runs the gauntlet, and the last gate is always a human.
We publish the failures too: the paper trader ran flat vs SPY — we say so. 0 of 118 strategy variants survived the cost gate — we wrote it down instead of selling it.
04 what it plugs into
Wired into the tools real work runs on.
The engine reaches your apps through the same connectors a developer uses (MCP) — so it doesn't just talk, it acts: opens PRs, files issues, drives a browser, pulls live docs.
05 the honest record
We publish what didn't work.
Most of building real AI is dead ends. We write them down instead of selling around them — it's the whole reason to trust what we do ship.
strategy variants cleared the cost gate in our market-edge hunt. The honest answer was "no real edge" — so that's what we wrote down, for $10.20 of spend.
the paper trader came out even with the market, not ahead. We report it flat, because it was — no cherry-picked window.
the audit caught a bug mis-grading 68% of betting picks as fake losses, then applied 496 corrections — before any of it counted.
06 the engine
Step inside the engine.
The full deep-dive — the architecture, the vault, the deterministic router, the model lineup and the audit pipeline, laid out the way an engineer would want it.