Questions and answers
What companies ask most often when they are planning to modernise a system or develop with AI.
Won’t the AI be trained on our code?
No. I only use commercial AI accounts, where under the providers’ terms customer data is not used for training by default. I don’t use personal or free versions of AI tools for client work. Details and links to the providers’ terms are on the Confidentiality and data protection page.
Do we need our own company AI?
No, I bring my own. Each client gets a separate workspace with its own key and spending limit. If you want everything under your control, the model can run in your own Azure in a European region, with the contract with Microsoft in your name.
Where does our data go?
Only source code and documentation go to AI, never production data such as customers, invoices or personal data. I test on synthetic or anonymised data. The code stays in your repository, and I delete all copies when the engagement ends.
What if we don’t want any cloud?
It can be done with an open-weight model on your hardware or mine. It is weaker and slower, so the work takes longer and costs more. I will tell you up front so that you can decide.
Who is responsible for code written by AI?
I am. AI is a tool. I review, test and stand behind every change just as if I had written it by hand. What to watch out for is described in What the AI agent got wrong.
Who owns the code?
You do. Code, documentation and access are yours once paid, without restriction.
Can an old system be modernised step by step, without a “big bang”?
Yes, and it is usually the better way. First I capture how today’s system behaves, then I move it over piece by piece, and each piece must give the same results as the original. Users keep working in the meantime. The Payment matching with AI case study shows what this looks like on a real example from Delphi and Oracle PL/SQL.
Do you work for large companies and public institutions?
Yes, I have spent most of my career working for banks, insurers and large organisations. I know their processes, security requirements and pace.
What does it cost, and how do we start?
It depends on the scope. We start with a free call about what the system does and what is bothering you. Within two business days you then get a fixed-price proposal for a concrete deliverable. The easiest first step is a mini-diagnosis, from about CZK 10,000. For long-term work we agree on a day rate.
What if the result is not as agreed?
Before we start, we agree in writing on what I will deliver, by when, for how much and how you will accept the result. I invoice for a delivered result that meets those criteria. If the brief changes along the way, we agree on the change before I start working on it.
Do you also work without AI?
Yes, if you want it that way or your rules require it. But to be honest, without AI the work is slower and more expensive. I explain why in Development speed: what changed in just a few years.
What does AI cost in a running application?
It depends on how much the AI decides. With a good design, the cost of the model is small next to the cost of human work. In Payment matching with AI it comes to about $0.36 per 1,000 payments, because the model is called only where it can change something. I will prepare an estimate for your case as part of the analysis.
Do you work on-site or remotely?
Mostly remotely; I am happy to come in for important meetings. I am based in Prague.
Is permanent employment also possible?
For the right team and company, yes. Permanent employment is one form of working together, just like a long-term contract or a project.
How do we work together?
- A free call: what you have and what you need.
- A proposal with scope, price and the terms for using AI.
- Work in phases. At the end of each phase you see the result and decide on the next one.
- Handover: code in your repository, documentation and recorded decisions.
Didn’t find your question? Write to me.
Want to know what AI could do with your system?
Let’s discuss it on a free call.