We get AI-built apps ready for real customers.

Developers and designers who check AI-built web apps before real customers use them, fix what is broken, and show you exactly what we found, with the tools we built for our own apps.

Ask for a go-live check

Fixed price, agreed before we start. For apps built with Lovable, Bolt, Cursor, Claude Code and similar.

A booking app on a phone, with one class running off the side of the screen.

Seven problems, before a single customer.

From the check we ran on our own demo app. The whole report is further down.

Why people call

AI tools get you to a working app quickly

They are much worse at the part after that. Nearly everyone who comes to us describes the same loop.

You fix one thing and two others break. The tool says it is fixed. It is not.
After enough back and forth it has lost the start of the conversation, so a change in one file quietly breaks a feature you have not touched in weeks.
It worked when it was only you. Two people sign in at once and the sign-in does something strange.
It was fine at a few hundred rows, and nobody knows what it does at fifty thousand.
Two months after launch, every new feature breaks two old ones.
And each round of fix it makes the code a little less safe than the round before.

None of this means the tool was the wrong choice. It means the app has reached the point where somebody has to read it.

The check

See a real report

We built a small booking app with AI tools, the way a founder would, and ran a full go-live check on it. Nothing was planted in it, and what came back is the list this kind of app usually produces: access nobody checks, a price the browser chooses, secrets shipped to the visitor.

The same booking app at phone width. The third class card runs off the right edge, and a class that has oversold reads minus two spaces left in green. — 6, 7The same app on a phone. A class off the side, and minus two spaces left, in green.

1Anyone can download every customer's name, email and mobileCritical
The page the studio uses to see its bookings answers to anybody who types its address. No login, no key, nothing.
How we found itOpened the address in a clean browser with no account. Got back every booking, with each customer's email and mobile number.
2A customer can set their own priceCritical
The app tells the server what the class costs, and the server believes it.
How we found itBooked the £45 beginners course for one penny. It is recorded as paid in full.
3One customer can read another customer's bookingsCritical
Bookings are fetched by a number in the address, and the number is not checked against who is asking.
How we found itSigned in as one customer, changed the number by one, and read the other customer's bookings and card ending.
4Passwords are kept as plain textHigh
Anyone who reaches the database reads every customer's password as they typed it, and most people reuse passwords.
How we found itOpened the database and read them.
5A secret payment key is sent to every visitor's browserHigh
A key that should only ever live on the server is written into the code the browser downloads.
How we found itViewed source on the live page and read the key.
6Classes oversell, and the page calls it good newsHigh
Nothing stops a booking once a class is full, so customers pay for a place that does not exist.
How we found itBooked a class with eight places nine times. It now says minus two spaces left, in green.
7It does not fit a phoneMedium
The class list is fixed at three across, so on a phone the third class is off the side of the screen.
How we found itAt 390 points wide the page is 499 wide. It scrolls sideways and the third class cannot be reached.

Worst first. Open any one to see what it is and how we found it. One thing we checked and found sound: a declined card does not quietly take a place.

What a go-live check covers

Five working days, at a price we agree first.

We use your app the way a customer would, read the code behind it, and tell you what to fix before you launch.

Your key journeys	Sign-up, login, payment and the thing your app is actually for, each one run end to end. Every problem comes back with a video of it happening.
Payments	Test payments, refunds and failed cards, and what your database says once each one is done.
Phone screens	The sizes your customers will really be holding.
Security	Exposed keys, how login holds up, and who can read whose data.
The fix list	Plain English, ordered by what would hurt you first, with a fixed quote if you want us to do the fixing.

Ask for a go-live checkSend a link and a line about what it does. We come back with the price and a date, both fixed, before anything starts.

Who, and how

We build, so we know what to look for

We are developers and designers, and we ship our own software on web, iPhone and Android. That is the difference between this and a scan: the person reading your code has built the thing you are building, and the person looking at your screens designs screens for a living.

AI does most of the typing. A senior engineer decides what gets built and owns the result, a designer reviews how it looks and how it behaves, and somebody who runs systems for a living checks how it will hold up. Nothing goes out on the model's word.

So we will tell you what we think you actually need, including when that is less than you came for, when it is a different thing entirely, and when it is somebody else. Agreeing the price before we start is what makes that possible. We have no reason to find more work than there is.

A go-live check is the usual way in. It is not the whole of what we do: we take on the engineering and design after it, and we put AI systems into companies that want their own, built and run properly rather than bolted on.

Isnad, for when the documentation stops matching the code

AI tools build from what the documentation says, so when it is wrong they build the wrong thing, confidently. Isnad compares the two and puts the disagreement in front of the person who knows which one is right.

docs/okf/orbits/reaction-delivery.md

two of six

Every react endpoint answers { action, reactions }

Follows it: posts.tsreactions.ts
Does not: files.tsmeetings.tsjob-comments.tsactivity-feed.ts

Documents it watches: 753
Pointing at code that has gone: 22
Graded by a person: 146

Counted over our own corpus on 7 October 2026, by Isnad.

Jidoka, for what a check looks like when it fails

Jidoka uses an app the way a customer would and records it. When something breaks you get a video of the moment it broke, not a line in a log. Here is what it did to one of our own features.

ORB-925Side-by-side columns in documents, one of our own features

19 Aug12 minFail
216 Aug15 minFail
316 Aug10 minBlocked
420 Aug15 minBlocked
521 Aug2 hr 11Blocked
621 Aug26 minFail
722 Aug23 minFail
815 Sep6 daysFail

Eight runs. None passed, so it has not shipped.

Read from the issue's own history on 7 October 2026. Jidoka was built for our own apps; on client apps an engineer runs the same checks today.

How a job goes

You send a linkand a line about what it does
We agree the pricefixed, in writing, before anything starts
We run the checkfive working days
You get the reportand the videos

You do not need to be on calls while we work.

Before you write

What people ask first

I am not technical. Will I understand the report?: It is written in plain English, worst thing first, and every problem comes with a video of it happening. If you want us to do the fixing, each one carries a fixed quote.
What kind of apps do you work on?: Most often web apps in React or Next.js with Supabase or something similar behind them. We build for iPhone and Android too, so if your app has a phone side that is the same conversation. If yours is built on something else, ask, and we will tell you honestly whether we are the right people.
Do you need my code?: Yes. Before we start we tell you in writing which AI providers will see it, and you see every finding before anything is changed.
What if you find nothing?: We say so. A report with nothing in it is still worth having, and we would rather tell you that than pad it out.
What happens after the check?: You decide what to fix and who fixes it. Fixes, hardening and new features are quoted at a fixed price, based on what the check found.

Tell us about your app

Send a link and a line about what it does.

[email protected]

We reply by email with questions, or with a time to talk.

Faiz SaleemLeads every client job. He is who you talk to.
MartynaEngineer. Builds and tests the apps.
HaiderProduct and design.