AI coding that holds up after the prototype

Three typical situations

You built an app with AI and want to go live

It runs, but you do not know whether it is safe. Before launch I check what really counts: access rights, secrets in the code, input validation, backups and costs under load. You can go through the main points yourself first with the pre-launch checklist.

The project is stuck

Every change breaks something else and the AI goes round in circles. Tests and a clear structure are often missing. I look at what can be saved and what should be rebuilt, and tell you plainly.

Your team should develop faster with AI

Rolling out tools is easy. The rules are harder: what may go into a prompt, who checks what, how does the code stay maintainable? Rolling out AI coding in a team covers what matters.

Why caution makes sense

The research supports a sober approach. In a 2025 randomised METR study, experienced open-source developers took longer with AI tools but believed they were faster. METR has since said this picture has probably shifted. Veracode found a known security flaw in about 45% of tested tasks in 2025. AI is a strong tool, but it does not replace checking.

How I work

I am a certified application developer, have built software for clients since 2012 and use AI coding tools in my own daily work. So I know where they save hours and where they write convincingly wrong code. You do not get slides, you get a prioritised list and, if you want, the fixes.

What you have at the end

• An honest verdict: launch, repair or rebuild
• Critical problems fixed rather than a long list of defects
• Tests for the flows that must not break
• Rules that let you or your team carry on alone

Related articles

• Vibe coding: a pre-launch checklistBuilt with AI and about to launch? This checklist covers access rights, secrets, data, costs and legal basics before real users arrive.
• Rolling out AI coding in a teamHow to introduce AI coding tools to a development team: data, reviews, tests and measurement, plus what the research really says on productivity.

Frequently asked questions

Do I have to be able to code?

No. Many people who build with AI do not come from development. I explain the findings so that you can decide.

Which tools do you use?

That depends on the project and on data protection. The process matters more than the tool, because tools change quickly.

Are my data and code safe?

Before a tool gets chosen, it is settled what it may see. Sensitive data does not belong in services without a suitable data processing agreement.