How I work with AI
I use AI every day. It makes me faster, and it makes some work possible that would not fit into a project otherwise. But it does not change who is responsible, and it does not change who decides.
In short
Three rules hold together. Responsibility stays with the person who made the decision: AI signs nothing. The architect decides the architecture: AI writes faster, but what to build and how the parts connect is a human decision. Every AI result is checked: nothing it produces moves on without a person reviewing it. In client projects the AI agent gets no access to the code, databases, connection strings, keys or passwords, and nothing enters the code that a person does not know about.
Three rules
1. Responsibility stays with the person who made the decision
When something fails in production, the question is who made the decision. The answer is never the model. If I put a solution into a project, it is my solution, whoever drafted it.
2. The architect decides the architecture
AI can write a function in seconds. What it cannot do is know the business, the constraints that are not written down, and the cost of a wrong decision three years from now. Choosing what to build, what to build it from, and how the pieces connect stays with a person.
3. Every result is checked
A suggestion from AI is treated like code from a new team member: useful, often right, and never merged unread. The check is not a formality; it is where most of the value of my work sits.
What AI does in my work
- Code suggestions. AI proposes a solution to a task set by a person. A person reviews it and puts it into the project.
- Reading other people's and old code. It reads a system quickly and proposes a description, which is then checked against real data and with the people who use the system.
- Drafts of specifications and documentation. A first version of the text, which a person edits and approves.
- A second opinion. Looking for mistakes and weak spots in a finished solution, and comparing alternatives.
- A QA report after a human wrote the code. A separate pass that looks for what the author missed. It adds to the review; it does not replace it.
- A first UX and UI proposal. A starting point for the initial version of an interface, which a person then shapes.
What I never hand to AI
- Architectural decisions. What to build, what to build it from, and how the parts of the system connect.
- Releases to production. Nothing goes to the live environment without a person deciding and checking.
- Client data. The data of my clients and their users is not passed to AI.
- Promises to the client. Estimates, timelines and audit conclusions are written and signed by a person.
Access
In client projects, the AI agent gets no access to the code, databases, connection strings, keys or passwords. Nothing enters the code that a person does not know about.
This is not distrust of a specific tool. It is the same principle as least privilege for any system: the agent does not need that access to make a suggestion, so it does not get it.
Frequently asked questions
Do you use AI in client projects?
Yes, every day: for code suggestions, reading existing code, drafts of specifications and documentation, a second opinion, a QA pass after code written by a human, and a first UX and UI proposal. A person reviews every result.
Does AI get access to my code or data?
No. In client projects, the AI agent gets no access to the code, databases, connection strings, keys or passwords, and client data is not passed to AI.
Who is responsible for code that AI helped write?
The person who decided to put it into the project. AI signs nothing; responsibility stays with the person who made the decision.
What do you never hand to AI?
Architectural decisions, releases to production, client data, and promises to the client such as estimates, timelines and audit conclusions.