# James Brady
> James Brady builds AI systems that show their work. One person, operating at fleet scale — documenting what actually works.

## Work with me
- [Work with me](https://www.jamesbrady.org/work-with-me): Work with me splits into three engagements: getting a business found in search and inside AI answers, delivered by the agency New Reward; building a system that ships in verified waves, delivered by the studio Utlyze; and background screening for new hires, delivered by Vuplicity. Each page states what the engagement measures or produces, who delivers it, and the budget band it normally sits in.
- [Get found in Google and in AI answers](https://www.jamesbrady.org/work-with-me/get-found): Getting found means two things now: ranking in an ordinary Google search, and being the source an AI assistant answers from. New Reward, the agency James Brady operates, measures both from outside a business, scores what it finds against a fixed rubric of 12 axes and 76 measures, and reports which gaps are worth closing first. Missing evidence is written down as missing. No outcome figure from a client engagement is published on this site yet. — delivered by New Reward
- [Build a system that shows its work](https://www.jamesbrady.org/work-with-me/build-a-system): Building a system, here, means a scoped engagement that ships in verified waves: a written scope, work landed in waves that each carry their own evidence, an evidence packet a non-technical owner can read, and a handoff where the client owns the repository, the checks and the deployment. Utlyze, the studio James Brady operates, does this work. Nothing is claimed as done until the place it lives has been read back. — delivered by Utlyze
- [Background screening for your hires, done clearly](https://www.jamesbrady.org/work-with-me/background-screening): Background screening means checking a person's record before hiring them, and the checks behind this page are run by Vuplicity. Vuplicity's public site describes nationwide background screening with clear pricing and cleaner workflows, publishes packages named Basic, Essential and Complete, and keeps separate pages for security, candidate consent and common questions. An enquiry started here reaches James Brady, and Vuplicity delivers the screening under its own published terms. No screening figure from any employer is published on this site. — delivered by Vuplicity

## Work
- [Work index](https://www.jamesbrady.org/work): The work index lists products, open-source projects, client work and experiments, filtered in the browser with a count read from what is on screen.
- [OfOne](https://www.jamesbrady.org/work/ofone): OfOne is an open-source decision compiler that turns a hard question into a typed, checkable map before it writes any prose. The map holds evidence, claims, unknowns, kill tests, causal edges, options, triggers and human gates. A validator program checks the map, and only a map that passes gets rendered into an answer. When a new fact arrives, only the chain that depended on the changed fact is patched, and the system reports what it invalidated.
- [plimsoll](https://www.jamesbrady.org/work/plimsoll): Plimsoll is a local-first collector that ties AI coding spend to shipped outcomes, so a team can say what its tokens actually bought. The collector watches Claude Code and Codex on your own machine, records what each session cost, and joins those sessions to merged pull requests and passing checks. Content is discarded before anything reaches disk and identifying strings are hashed. The published unit is cost per merged pull request.
- [The visibility platform](https://www.jamesbrady.org/work/visibility-platform): The visibility platform measures how findable a business is in ordinary search and inside AI assistants, then scores it against a fixed rubric of 12 axes and 76 measures. A fixed rubric is what makes this month comparable to last month, and one business comparable to another. Every client is a tenant with walled data. Missing evidence is recorded as missing and never scored as a zero, because a report that fills gaps with zeros misleads the person paying for it.
- [EEG meditation toolkit](https://www.jamesbrady.org/work/eeg-meditation-analysis): The EEG meditation toolkit is Python tooling that turns a raw consumer-headset recording into a readable account of a meditation session. Filtering removes artifacts, power is measured in each brainwave band, and a depth score from 0 to 100 rolls up with one of four labels: alert, light, moderate or deep. A synthetic data generator ships with it, so the whole pipeline runs without a headset and without handing over a personal recording.
- [seopr1.com](https://www.jamesbrady.org/work/seopr1): seopr1.com is a five-page marketing site for AI-search visibility, built so the site itself survives the scrutiny the service invites. Pages ship as static HTML with one framework dependency; a WebGL tuning fork tied to scroll carries the hero, and every effect has an explicit fallback. If WebGL is missing, the whole DOM experience remains and only the 3D drops. Nothing important lives inside the animation.
- [AI-readiness assessment](https://www.jamesbrady.org/work/ai-readiness-assessment): The AI-readiness assessment scores an organization across seven competency domains using 28 questions, and returns four views of the same result rather than a single number. Questions filter by role, so an executive, a manager, an IT lead and an HR lead each answer what they can actually judge. Four views matter because one number hides which two domains are dragging the average down. Results export as a formatted PDF or as raw JSON.
- [The Of One family](https://www.jamesbrady.org/work/of-one-family): The Of One family is a thesis being tested, not a finished product line: a governed set of domains and applications that each name a single company role and ask what that role looks like as one person plus AI. Two of them are live applications. The rest is staked ground: domains and landing pages that exist to find which role-shaped problem pulls interest before anything gets built. The page labels which is which, every time.

## Theories
- [Theories index](https://www.jamesbrady.org/theories): The theories index lists open questions worked in public, with each theory's maturity state shown as prominently as its title.
- [Why do questions have geometry?](https://www.jamesbrady.org/theories/universal-question-geometry): Universal Question Geometry treats a question as an operation that changes the state of knowledge, which means questions can be ranked and scored before any of them are asked. The method keeps a running picture of the current state of knowledge, estimates how much each candidate question would move that picture, and selects the next question from the top of that ranking. Stopping is not allowed on the feeling of being finished; a stop attempt is rejected while a question with positive expected value is still open. — maturity: developed
- [Why should a useful-looking finding get more scrutiny, not less?](https://www.jamesbrady.org/theories/question-answer-dynamics): Question-Answer Dynamics is a rule for interrogating every finding before that finding is acted on, and the central claim is that scrutiny should scale with how convenient a finding is. A finding that contradicts the plan already attracts hard questions; a finding that flatters the plan does not, which is exactly why it slips through unchecked. The method fixes this by running a fixed battery of questions on every finding and by refusing to stop research on satisfaction. — maturity: developed
- [Can one person run a software team by splitting judgment from building?](https://www.jamesbrady.org/theories/architect-loop): The Architect Loop is a working arrangement in which one AI model does judgment only and a separate set of AI models does all the building, with a git repository as the only shared memory. Judgment means arbitration, verdicts against frozen gates, splitting work into lanes, and deciding whether to continue or kill. Building means writing code in isolated worktrees, with no authority to decide whether its own work passed. — maturity: developed
- [Do language models share a hidden vocabulary for relational states?](https://www.jamesbrady.org/theories/latent-emotions): Latent Emotions is a personal, unfinished observation that different large language models reach for closely similar words when describing certain relational states, which may point to shared structure underneath rather than shared training vocabulary. No claim of proof is made here, and no formal study exists. The work is paused, and the page says so instead of dressing an observation up as a programme. — maturity: sketched, paused
- [What is a five-layer operating system for an agent-run business?](https://www.jamesbrady.org/theories/the-paper): THE PAPER is a single working document that specifies a five-layer operating system for running a business where AI agents do most of the labour. The five layers are business truth, competitive monitoring and recommendation, strategy, execution, and visibility measurement, each one gated by the layer beneath it. The document is a specification under active revision, not a published theory, and this page introduces it without reproducing it. — maturity: sketched
- [What is a style rule a program can check?](https://www.jamesbrady.org/theories/movement-economy): Movement Economy is a rule that every sentence in a piece of analysis must do one of ten declared jobs or be deleted. The ten jobs are: bound the question, ground it in evidence, make a claim, link two things, state a test, propose a move, evaluate options, warn, set a trigger, and gate a decision. Anything that does none of these is decoration, and decoration is removed. — maturity: sketched
- [Should proof of function come before proof of thoroughness?](https://www.jamesbrady.org/theories/function-first-orchestration): Function-First Orchestration is an operating rule that puts proof of function ahead of proof of thoroughness. Under the rule, a piece of work ships with one honest happy-path test plus the specific refusals that protect money, custody, and truth, and nothing else. Broader test coverage is deliberately deferred rather than skipped. Each deferral is written into a parking file, so “later” becomes a concrete list entry instead of a vague intention. — maturity: sketched

## Lab
- [Lab](https://www.jamesbrady.org/lab): The lab holds interactive artifacts, each one shipping with a written explanation page; an artifact without one is excluded from search engines by the loader.
  - The emotional manifold: The emotional manifold is a visual demo that moves a conversation through a six-axis space (energy, valence, complexity, novelty, introspection and focus) instead of down a page. The six axes were arrived at in conversation with a language model rather than derived from anything, so the demo is a lens for looking at an idea and not evidence for it. The theory behind it is paused, and the demo is labelled accordingly.

## Learn
- [Learn hub](https://www.jamesbrady.org/learn): The learn hub introduces three archived volumes — the Primer, the Manuscript and the Workshop — each kept at its original URL with a dated archive badge.
- [The Primer](https://www.jamesbrady.org/primer): The Primer explains how modern coding systems fit together for a reader who does not write code: what code is, what a stack is, what an AI agent does, what a skill is, and what MCP connects. The volume is archived as written and kept at its original URL. Its prose was not rewritten when the page moved onto the current design; the archive date says when it stopped being maintained. — archived 2026-08-11
- [The Manuscript](https://www.jamesbrady.org/manuscript): The Manuscript is a curated catalog of tools and MCP servers, grouped by the job each one does, with an install line and a source link for every entry. The catalog count shown anywhere on this site is computed from the catalog source at build. An earlier version of this site printed a hand-typed figure that was wrong by a factor of eight, and that class of defect is what the computed count exists to prevent. — archived 2026-08-11
- [The Workshop](https://www.jamesbrady.org/workshop): The Workshop is three practical guides that end in a working setup: standing up an AI agent, installing a skill, and connecting an MCP server. Each guide is written to be followed once, start to finish, rather than skimmed. The volume is archived as written and kept at its original URL, with the archive date stated on the page. — archived 2026-08-11

## Site
- [James Brady — builds AI systems that show their work](https://www.jamesbrady.org/): James Brady builds AI systems that show their work, and every figure on this site is computed from its source at build rather than typed by hand.
- [About](https://www.jamesbrady.org/about): James Brady operates two entities: Utlyze, the studio, and New Reward, the agency. The about page also answers how one person checks work before it ships.
- [Research projects](https://www.jamesbrady.org/now): James Brady is currently working on a number of different research projects. Which projects those are is not published here yet, and what ships from them will appear on this site with its numbers and its method when it is ready. This page is hand-edited rather than generated from repository activity, the age indicator beside it is computed from the date the file itself carries, and the open questions the whole site is still carrying are listed below. — updated 2026-08-12
- [Contact](https://www.jamesbrady.org/contact): Contact routes a qualified enquiry to James Brady through the existing lead gateway, and states plainly what makes a project a strong fit.
- [Links](https://www.jamesbrady.org/links): The links page collects James Brady's public profiles, every way to work with him, and the recorded walkthroughs, in one place at its original URL. Profiles listed: GitHub, LinkedIn, X, TikTok, YouTube, Bluesky, and a direct email address.
- [Watch](https://www.jamesbrady.org/watch): The watch page holds recorded walkthroughs, frozen with the archive and kept at its original URL.

## Machine-readable
- https://www.jamesbrady.org/feed.xml
- https://www.jamesbrady.org/.well-known/ai-manifest.json
- https://www.jamesbrady.org/sitemap.xml
- https://www.jamesbrady.org/api/catalog

## Terms
Content © James Brady. Cite the source URL. Figures are computed at build; method is stated on-page.