AI Sign-Off is a developer review and sign-off for software your business made with AI: we check it against what it is meant to do, fix what fails, and then take accountability for it, in writing for the version we reviewed and on an ongoing basis if you keep us on a support retainer.
It is one of our development services, and it exists because when a tool was built with AI, it is often unclear who stands behind it.
Close, Not Correct
AI tools are good at producing something that runs. They are less reliable at producing something that is right in every case.
You describe an app, a booking form or an automation in plain English and a few hours later you have something that runs. The part nobody can see from the outside is which bits are exactly right and which are only close. A login screen that looks finished may not check whose data a user is asking for. An automation that works on the ten examples you tried may do something odd on the eleventh. The risk is in the parts that are only close, and you can’t see them until a customer finds them.
AI Sign-Off puts a developer between that output and your customers: someone who reads code for a living checks it and puts their name to the result.
Someone Has to Stand Behind It
Somebody has to answer for the software your customers use. When a developer wrote it, that answer was obvious. When it came out of a chat window, the answer is uncomfortable: the tool takes no responsibility at all, and the person who typed the prompt often can’t vouch for code they never read. Our article on who is responsible for software built with AI goes into that properly.
A sign-off puts a name to it: a named developer, a named version and a stated scope.
Who It Is For
- Owners who built an app, internal tool or website feature themselves with AI, the practice usually called vibe coding, whether that was in ChatGPT, Claude, Cursor, Lovable, Bolt or Replit.
- Teams where staff have built automations or AI agents that now touch customers, money or data.
- Businesses that had something built quickly with AI by someone else, and now have nobody who can vouch for how it works.
It is not for production systems with an existing development team that already reviews its own code. And it is not for anyone who wants a rubber stamp. We read the code, or we don’t sign.
What We Check
The sign-off is always against a stated scope. These are the areas that scope covers:
- Does it do what it’s meant to do. We test it against a short written brief of intended behaviour. If none exists, we write it with you first.
- Security. Keys and passwords sitting in the code or the browser, how logins and permissions work, how it handles what users type in, admin areas left open, and third-party packages with known flaws.
- Data. What personal or business data it touches, where that data is stored and who can see it, checked against the UK GDPR basics.
- Failure. What happens when something goes wrong: whether errors are handled, whether anything is logged, and whether anyone would notice.
- Running costs and limits. AI and API usage that could run up a bill, rate limits from the services it depends on, and the hosting it sits on.
- Maintainability. Whether another developer could understand it and change it safely. That means structure, tests and version control.
- Agents and automations. What it is allowed to do on its own, where a person must approve an action, and how it hands over to a human when it is unsure.
How It Works
- Brief. We agree in writing what it should do, who uses it and what data it touches.
- Review. A developer reads the code and tests it against the brief and the checks above.
- Fix. What fails gets fixed. Small items are fixed within the review day; anything larger is quoted before we start.
- Sign off. You get a written sign-off of the reviewed version against the stated scope.
- Own. If you want it, a support retainer keeps us accountable for the code as it changes.
Say an office manager built a customer booking tool over a few weekends with an AI coding assistant. Staff use it daily and it now stores customer names and phone numbers. A review of something like that could easily find a key for the AI service visible in the browser and an admin page that opens without a login. Neither takes long to put right, so both are fixed on the day. The sign-off then names that exact version and says what was checked.
What You Get
- A written review in plain English: what we checked, what passed, what failed, and what each finding means for your business.
- The fixes, either made during the review or quoted before any larger work starts.
- A signed statement naming the exact version reviewed and the scope it was reviewed against.
- The option to have us own it from there on a monthly retainer.
The report opens with the verdict, so you know where you stand before reading any detail. It is one of three: signed off, signed off with conditions, or not yet. Under it, each of the seven areas above gets a pass, a condition or an open issue, with one line saying why. After that comes every finding, each with how serious it is, how long the fix takes and whether it has been done.
Here is how one finding reads, taken from the booking tool example above:
The admin page opened without a login · Must fix · 30 minutes · Fixed
What we found: the page listing every booking, with customer names and phone numbers, opened for anyone who knew its address.
What it means for your business: anyone who found the address could read your customers’ details, which is personal data you are responsible for under UK GDPR.
The fix: the page now asks for a staff login on every visit, however someone arrives. Re-tested in a private browser window.
What Moves the Price
AI Sign-Off starts from £750 + VAT. That covers one day of review for a small app or automation, including the small fixes. Larger codebases are quoted after a first look, before you commit to anything.
The things that push a quote up are the size of the code, the number of outside services it connects to, agents that can act on money or customers without a person approving, and how much fails on the first pass. Access matters too. We need the code itself and a way to run it, because a review done from screenshots would be guesswork. Our guide to what it costs to fix a vibe-coded app explains when a fix is a patch, a refactor or a rebuild, and how to compare quotes.
Ongoing accountability sits on a support retainer. Our support retainer pricing page explains how those are structured.
What We Will Not Do
We will not sign off code we haven’t read, a version we haven’t seen, or anything outside the scope written into the sign-off. We won’t describe anything as fully secure either. The sign-off records what we checked and which version we checked it on.
We are also not accountable for changes made after the sign-off unless you are on a retainer with us. This matters more with AI-built software than with anything else, because the next change is usually one more prompt away. The moment someone asks the tool to add a feature, what is running is a new version that nobody has reviewed.
A sign-off is not a penetration test. If the stakes call for someone deliberately trying to break in, that is security testing, and we will tell you when it is worth it. If what you need is an independent opinion on whether to keep, fix or replace what you have, technical consulting is the better starting point.
A Brief You Can Send Anyone
Whoever you ask to review AI-built software, these five lines are what they need first:
What it does: [in a sentence or two, as you would explain it to a new member of staff]
Who uses it: [staff, customers, or both, and roughly how many]
What data it touches: [customer details, payments, internal records]
What it was built with: [the AI tools used, and who did the building]
Where it runs: [hosting, accounts, and who holds the logins]
Before You Get in Touch
If you would like to do a first pass yourself, our guide on how to get a vibe-coded app ready for real users walks through the checks an owner can run without a developer. For a quicker read, the AI-built app risk check asks ten questions and shows which areas need work. If you already use an automated AI review tool and are wondering whether that covers it, we compare the two in AI code review tools vs developer sign-off. The thinking behind the service is in our field note on why AI-built software needs a developer to read it.
Send us the five lines above, or as many of them as you can answer, and we will tell you whether it fits in a day or needs a quote. Get in touch.