You'll get a written report that shows what needs attention first.
If you want me to stay on after that, it's a monthly retainer that starts from those findings.
What the assessment covers
- Architecture:
- how the pieces fit, where your data goes, and what breaks first when more people use it.
- Code:
- whether someone other than the AI that wrote it can read it, change it and trust it.
- Security:
- who can see what, where keys and passwords live, and what a stranger can reach from the internet.
- Reliability:
- what happens when a service goes down, and whether a backup has ever been restored.
- Testing:
- which parts are tested, and which could break quietly on the next change.
- Deployment:
- how a change gets from your laptop to your users, and how you undo a bad one.
What you get
A written report in plain English, ranked by what to fix first. Each finding says what I found, what could go wrong for your users or the businesses you sell to, and how to fix it. Then we go through it together on a call.
| Priority | Finding | Risk | Fix |
|---|---|---|---|
| Fix now | The admin pages check for a login and never check for the admin role. | Any signed-in user who types the address can open them. | Check the role on the server for every admin page. |
| Fix now | A payment API key is saved in the code repository. | Anyone with access to the repository, now or later, can use it. | Move it to the host's secret settings and issue a new key. |
| Before launch | Backups run every night and have never been restored. | Nobody knows they work until the day one is needed. | Restore one to a test database and write down the steps. |
| Next | Sign-up has no automated tests. | A change somewhere else can break sign-up and nobody notices. | Add a few tests that run on every change. |
I'm Jared Irish
I'm an AI solutions architect, a computer engineer and a data scientist. I spent 24 years in the federal government. I founded NotJustLuck.ai and I'm based in Charleston, SC. I also write Claude Code skills.
Every day I use Claude Code and Codex for my own products and client work. I review each change before it ships.
Where code from ChatGPT and Claude tends to slip
I build with these tools every day, so I know their habits. The code usually runs. The gaps are in the parts you can't see from the screen:
- API keys or passwords sitting in the code
- a page that checks you're logged in but never checks whether that record is yours to see
- one database shared by testing and real customers
- backups nobody has tried to restore
- error messages that show users the app's internals
- the same rule written three different ways in three files
How it works
- A short call. You tell me what the app does and what's coming up for it.
- A short letter with the scope and the fee. We both sign it before I look at any code.
- Read-only access to the code and the hosting account. I don't change anything during the review.
- The written report, and a call to go through it.
- If you want help with the fixes, we scope that as its own project, or as a retainer.
AI tools and your code
I use Claude in my own work. Before we start, I'll tell you where it touches your code, and that goes in the letter.
Support after the assessment
The report's priorities are the starting point. I can fix them as a separate project, or stay on as your technical person on a monthly retainer: reviewing changes before they go live, building new features, and joining the call when a customer's IT team has questions. I don't take on open-ended maintenance outside a retainer.
Email me about a retainerTell me about your app
Tell me what you've built and what you want reviewed. If you want me on retainer after the assessment, say so.
Email me about your appLooking for help with AI workflows instead? See what I build.