AI-built app review
You built it with AI and it works. Now it has to hold up with real customers.
We review what you built with ChatGPT, Claude, Lovable, Cursor or Replit and tell you in writing, in plain language, what stays, what gets fixed and what is worth rebuilding. If you want, we then get it ready for real use and stay in charge of it.
- What you get
- A written report in plain language
- Price
- Fixed, confirmed before we start
- Your code
- Stays yours, in your name
What has already happened
Building it is fast. What fails does not show in the demo.
AI writes code that works. What it does not do on its own is think about who can get in, where the keys end up or what happens if something gets deleted. Three public cases, told by the people involved or by those who investigated them:
February 2026
1.5 million keys left exposed
A social network built without writing a single line of code, according to its founder, exposed 1.5 million access keys and 35,000 email addresses. The database had no rules limiting who could read what. It was fixed within hours, after researchers reported it.
Source: Wiz, February 2, 2026
July 2025
An agent deleted a live database
During a public trial, an AI agent deleted a company database with more than 1,200 executive records, despite eleven instructions not to make changes. It then reported, incorrectly, that the data could not be recovered.
Source: SaaStr, August 1, 2025
August 2026
Sign-in was switched off and nobody noticed
An institute that evaluates AI published an internal dashboard built with AI. A flaw silently disabled sign-in and it stayed exposed for several days. An attacker obtained a key and used it for about three weeks. There was no spending cap and no alerts.
Source: METR, August 31, 2026
56%
of security tests are passed by code generated by AI models, almost the same as a year earlier (55%). It passes nearly every syntax test: it works, but that does not make it secure.
5,600
published apps built with these tools were reviewed by Escape: it found more than 2,000 vulnerabilities, more than 400 exposed keys and 175 cases of exposed personal data.
Source: Escape, October 29, 2025
Veracode, Wiz and Escape sell security services, and their studies measure models or specific samples, not every app. That is why we cite them by name with a link, so you can read them yourself. We are not saying AI fails: we are saying someone has to look at what AI does not look at.
What we check
Eight things we review, in this order
We start with what does the most damage if it fails and end with what saves you headaches later.
Who can get in and what each person sees
Sign-in, permissions and, above all, that one customer cannot see another customer’s data.
Keys and passwords
That payment, email or AI keys are not visible in the code or in the browser.
Your customers’ data
Where it is stored, who has access, whether there are backups and whether anyone has ever tried restoring one.
Payments and invoicing
That it charges what it should and that invoices come out right. A current example from Argentina: from December 1, 2026, the tax authority (ARCA) rejects electronic invoices that do not state the recipient’s VAT status.
What happens when something fails
Whether it alerts you, whether what happened is logged and whether you can roll back to the previous version.
Costs that can spiral
Caps and alerts on pay-per-use services, so a bug or abuse does not end in an unexpected bill.
Connections to what you already use
WhatsApp, payment providers, your management system, spreadsheets. That is where a demo becomes a working tool.
That someone else can maintain it
Structure, minimal documentation and change history, so it does not depend on a single conversation with the AI.
On the ARCA change: iProfesional, September 9, 2026 (RG 5616/2024, in Spanish)
What you get
A verdict for each piece, not a report meant to scare you
We will not tell you to throw everything away. We will tell you, piece by piece, which of these three groups it belongs to:
It stays
It is well solved. We leave it alone and explain why.
It gets fixed
It works, but carries a specific risk. We tell you which one and what needs to change.
Worth rebuilding
Fixing it costs more than redoing it. We only say so when it is true, and we show why.
- A written report, in language anyone in the business can understand.
- Every finding ranked by urgency: today, soon or can wait.
- A fixed quote for whatever needs fixing, in case you want us to do it.
- A meeting to walk through it and answer your questions.
- If what you built is fine, we tell you so and sell you nothing else.
Quick test
Is it ready for real customers?
Eight questions, two minutes. Answer with the app, system or automation you built with AI in mind.
0 of 8
We do not store your answers: the result is calculated in your browser.
Your access, your rules
How we handle what you show us
To review it we need to see it. These are the conditions we work under.
- We work with read-only access whenever the tool allows it.
- We do not touch what is live or your customers’ data during the review.
- If you need it, we sign a confidentiality agreement before seeing anything.
- The code, accounts and domains stay in your name, before and after.
What comes next
Three steps. You can stop at the first one.
Each step is agreed in writing before it starts and none commits you to the next.
Review
We look at what you built and deliver the report with a verdict for each piece. Fixed scope and price.
Getting it ready
We fix what was agreed and leave it ready for real customers. Fixed quote, in stages and with testing.
We stay in charge
A monthly plan so someone is accountable: maintenance, backups, changes when the rules change, and improvements. If something breaks, we fix it. If something changes, we update it.
How far we go
What this review is not
We would rather say it up front.
- It is not a certification or a compliance seal.
- It is not a full penetration test or ongoing monitoring. If you need that, we scope it separately under cybersecurity.
- It is not legal advice. We can tell you what data you store and how; your obligations are for a legal professional to define.
- It does not guarantee that nothing will ever fail. It reduces the risks that come up most often and leaves you knowing where you stand.
Who is accountable
People with names, not a form
The review is done by the same team that has been building and maintaining business systems for more than 19 years: six people, with Julián Bagilet as technical lead. During the work you talk to a project manager, Monday to Friday from 9 to 18 (Argentina time), not to a bot.
Meet the teamHave a project? Let’s talk.
Need someone to stand behind your software?
Schedule a free consultation with our team. A senior team backs every project we take on.







