When I say that we build AI systems, people nod. When I say that I take companies and teams into the world of AI Driven Development and vibe coding, they ask: What does that mean concretely?
The misunderstanding
Most companies believe that introducing AI is a technology project. You buy licences, you set up a coding agent, you run a training session. Done.
Then this happens: Two developers use it intensively and become fast. Five try it out and drop it again. The rest carry on as before. After three months the CTO says productivity has not measurably increased. And he is right.
What I actually do
I join a team that builds software. Banks, telecommunications, industry. People who know their craft. And I work with them so that they no longer write every line themselves, but lead agents that do it for them.
That sounds like less work. It is different work. Instead of writing, you have to describe. Instead of debugging, you have to check. Instead of building a solution, you have to decide which of three generated ones is the right one, and why.
A senior developer who understands this becomes very much faster within a few weeks. One who does not understand it gets more code and more problems.
How an introduction like this runs
I have done this often enough by now to see a pattern.
Weeks 1 to 2: Watching. How does the team work today? Where does the time go? What rules exist, and which ones live only in the heads of two people?
Weeks 3 to 4: A real ticket, with an agent, together. No example, no toy project. Something from the backlog that has to be done anyway.
Weeks 5 to 8: Writing down the rules. Architecture, conventions, what the agent may do and what not. This is the part everyone underestimates.
Afterwards: The team works on its own. I come back at intervals and look at what has crept in.
The most important step is the third. If a team does not write down its rules, the agent works by its own. And those never fit a system that has grown over fifteen years.
Why NeoCoder helps with it
We built NeoCoder for ourselves first, because we could no longer keep our fifty micro-models and our own projects under control any other way. By now I take it into teams because it enforces exactly the structure a team does not yet have at the start.
An architect agent that writes the specification and contract first. Coding agents that only work against this contract. A QA that is not the agent itself. Artefacts instead of chat histories. These are not features. These are the rules a good team would have anyway, except that now they are no longer optional.
For a team that is just starting out, this is an enormous relief. It does not have to figure out on its own how to stop agents from building things into each other's code. We have already figured that out, at our own expense.
About the resistance
In every team there is someone who does not want to. Usually it is not the weakest, but one of the best. Someone who is proud of their code, and rightly so.
I do not try to persuade this person. I give them the hardest task. Writing the rules. Deciding which patterns apply where. Showing the agents how this system works. That is architecture work, and usually after a few weeks the person realises that this is exactly what they always wanted and never had time for.
What is different afterwards
Fact is: the teams get smaller. I say that openly, even if nobody likes to hear it. Where you used to have twenty people, five good ones with agents are enough.
But the five work differently. They talk more about architecture and less about syntax. They argue about contracts instead of indentation. They know what the system costs, because they see every agent call. And for the first time in years, they have time for the questions that are really hard.
That is the world I take teams into. Not the world where AI does everything. The world where people once again do what they were hired for.
Anyone who wants to work with me gets no training course and no tool. They get a few weeks in which their team learns to lead code instead of writing it. After that they no longer need me. That is how it should be.
Written by me. The thoughts, the values, the learnings, the mistakes: all mine. Grammar and spelling are corrected by our own twin model, trained on my texts. Sometimes a stumble stays in. That is mine too.