Inspired By Frustration// Fixed price · 1–2 weeks · scoped on a 30-minute call
A fixed-price AI readiness assessment: where the AI spend stalls, what the agents may touch, which checks are missing before AI output reaches a customer, and a 90-day operating model your team can run without me.
Why teams hire us
Real, shipped production work behind every engagement — not advisory slideware or portfolio mockups.
11
client reports delivered from one pipeline
55
merged PRs/day, 30-day average
8 min
hard CI cap across 12 production repos
In short
An AI audit is a structured review of how an organisation actually uses AI: which models, assistants and agents are in production, what each one is allowed to touch, how its output is checked before it reaches a customer, and whether the spend is returning anything.
Most teams that call this in are not failing at AI.
They bought the licences, ran the pilot, watched the demo work, and then nothing reached a customer — pilot purgatory.
An AI audit is the cheap way to find out why, and whether the answer is worth more money.
It is deliberately not a compliance audit: a compliance audit asks whether the use is lawful and documented, and stops there.
This one asks whether the thing ships — what the agents may touch, which outputs are measured, what breaks silently, who signs off, and what the next ninety days should look like.
Fixed price, one to two weeks, scoped on a 30-minute call.
The verdict is written down, including the version where you do not need to hire anyone.
Every model, assistant, agent and licence in play, what each was bought to do, and where the work actually stops. Usually two or three of them are paying for the same thing twice.
A written boundary per system: which repos, databases and customer data each agent or assistant can read and write, what needs human sign-off, and what is off limits entirely.
Evals, CI gates and observability. Which AI output is measured before it reaches a customer, which is not, and what a regression would look like on the day it happens.
Access control, audit trail, and where the EU AI Act actually bites for your use case — written for the people who have to run it, not as a framework diagram.
Who owns which decision, how a new AI use case gets approved, what the review rhythm is, and how the team knows when to kill one. Small enough that it survives me leaving.
One document, ranked by what it costs you to leave alone — including the honest version where the answer is 'fix these three things yourselves and hire nobody'.
Four surfaces. What is running (models, assistants, agents, licences, and who pays for them). What they may touch (systems, data, write access, and where human sign-off sits). What is measured (evals, CI gates, observability, and the failure modes nobody has looked at). And what happens next (a 90-day operating model with named owners). Everything else — vendor politics, a maturity score out of five, a slide about transformation — is left out on purpose.
A policy says what is allowed. An audit says what is actually happening — and those two documents disagree more often than not. The common finding is not a policy breach; it is that nobody can say which assistant has write access to production data, or which AI-generated output reaches a customer without a human reading it first. That gap is what stalls the rollout, and it is not visible from the policy.
An AI readiness assessment asks whether an organisation can put AI into production at all: the data it can reach, the engineering practice around it, the decision rights, and the appetite for the failure modes. This engagement covers both — the readiness assessment is the first half, the audit of what is already running is the second — because for most teams the two questions arrive together. If you have shipped nothing yet, it is a readiness assessment. If you have shipped something and it stalled, it is an audit.
Whatever the verdict says, and often that is nothing further. When the finding is that the calls need making — architecture, vendors, hiring, governance — the fractional AI CTO seat is the follow-on. When the finding is that the calls are already made and there is a concrete AI surface to ship, the forward-deployed engineer engagement is. The audit is priced so it stands on its own either way; it is not a discovery phase that has to become a retainer.
You describe what you bought and what stalled. I come back with the scope and one fixed price for it — no day-rate meter, no discovery phase to buy first.
Codebase, prompts, agent configuration, access grants, spend, and the people who use it. Interviews with whoever actually runs the AI, not only whoever signed for it.
Findings ranked by what they cost you to leave alone, the boundary of what the agents may touch, the missing checks, and the 90-day operating model.
One walkthrough with the team that has to run it, the document in your hands, and no standing invoice. If a follow-on engagement makes sense the audit says so; if it does not, it says that too.
This is a productized audit, not a bespoke essay: the same report machinery that produced 11 client reports produces this one — fixed shape, fixed scope, fixed price.
Measured on the self-hosted Fly runner fleet, peak 111 in a day. The operating model the audit hands you is the one running here every day, not a template.
The fail-closed merge gate behind that throughput. The 'which checks are missing' half of the audit is written by the person who maintains this one.
Radiant Flow Yoga
Atlanta yoga studio — marketing & search audit delivered.
LeadingMomentum
Web platform designed, built, and shipped end to end.
Whiteleaf Consulting
M&A advisory — marketing site built and run.
Bring the pilot that stalled, the licence renewal you cannot justify, or the assistant nobody can say what it touches. Thirty minutes is enough to tell you whether an audit is worth it — and what it would cost.
The demo landed, the rollout did not, and nobody can name the blocker. Two weeks and a written verdict is cheaper than another quarter of the same.
Get a reply in 4hYou are signing AI invoices and taking the team's word on what they buy. If the finding is that the calls need an owner, the fractional AI CTO seat is the follow-on.
Fractional AI CTOYou already know the use case and the bar. Skip the audit — the embedded engineering engagement is the door you want.
Forward-deployed AI engineerAn AI audit is a structured review of what AI is actually running, what it may touch, how its output is checked, and whether the spend is returning anything.
A compliance audit asks whether the use is lawful and documented. An engineering audit asks whether it ships, and what is stopping it.
The common finding is not a policy breach — it is that nobody can say which assistant has write access to production data.
Fixed price, one to two weeks, scoped on a 30-minute call: the audit is built to stand alone, not to become a retainer.
The deliverable is a written verdict, a boundary for what the agents may touch, the missing CI and eval gates, and a 90-day operating model.
An engineering audit, not a compliance one
The report is written by the person who would have to fix the findings. Lawful-and-documented is table stakes; the question this answers is whether it ships.
The operating model is one he runs
Not a template. The review rhythm, the sign-off points and the CI gates in the 90-day model are the ones running across twelve production repos here every day.
Fixed price and a fixed end
One price agreed before the work starts, one to two weeks, one document. No day-rate meter, no discovery phase you have to buy first, no standing invoice afterwards.
// what clients say
We came in with a Lovable prototype and a board deadline. Three weeks later we had a typed backend, real auth, and an MCP server our support agents actually trust. The POC went to production without the usual rewrite tax.
DaanHead of Engineering
fintech scale-upPOC → production
I needed someone who could orchestrate a swarm of coding agents and still own the architecture. The agent-orchestration setup shipped 40+ PRs in a week — every one reviewed, scoped, and reversible. No hallucinated mess to clean up.
M.R.Founder
B2B SaaSagent orchestration at scale
The MCP integration was the part three other vendors quoted us six months for. Here it was live in under three weeks — tool schema, OAuth, rate limits, traces, the lot. Our Claude agents finally touch real data safely.
PriyaVP Product
healthtech startupMCP integration
One to two weeks, fixed price, scoped on a 30-minute call. You get a written verdict, a boundary for what the agents may touch, the missing checks, and a 90-day operating model — including the version where you hire nobody.
Scope the AI auditAn AI audit is a structured review of how an organisation actually uses AI: which models, assistants and agents are running, what each is allowed to touch, how its output is checked before it reaches a customer, and whether the spend is returning anything. A useful one ends in a written verdict with named owners and a plan for the next ninety days — not a maturity score.
It is a fixed price, agreed before the work starts, and scoped on a 30-minute call — the number depends on how many systems the AI touches and how much of it is already in production, so quoting a figure before that call would be guesswork. The engagement runs one to two weeks and ends with one document; there is no day rate and no discovery phase to buy first.
A compliance audit asks whether the use is lawful, documented and defensible to a regulator; a security audit asks whether an attacker can get in. An engineering AI audit asks whether the thing ships: what the agents may touch, which outputs are measured before a customer sees them, where a failure would go unnoticed, and why the pilot stalled. The three overlap on access control and audit trail, and answer different questions everywhere else.
An AI readiness assessment asks whether an organisation can put AI into production at all — the data it can reach and trust, the engineering practice around deployment and review, who holds the decision rights, and whether it can live with the failure modes. It is the forward-looking half of this engagement: readiness for what you have not built yet, an audit for what is already running.
Against evidence, not a questionnaire. Can you name every AI system in production and its owner? Can you say what each may read and write? Is there an eval that runs before a prompt or model change ships? Does anyone see the cost per task? Is there a human sign-off on output that reaches a customer, and is it real? A team that can answer those five with artefacts rather than intentions is ready; the gaps are the roadmap.