Two camps, one wrong question
Two camps have formed around AI-assisted software development. One says AI will soon write all the code and developers will become obsolete. The other says it's all hype and generated code is worthless in the end.
Both are wrong, because both ask the same question: how much can AI take over? The more interesting question is a different one: where does human judgment belong? We've worked with AI every day since ChatGPT went public in November 2022, today mainly with Claude Code. Over that time, the answer has become fairly clear to us.
format_quoteFor us, AI is a tool in experienced hands. Not a replacement for the hand.
Stefan Hess
Where AI speeds up the work
In our work, AI speeds up a lot of things in very concrete ways. A first draft of a new feature. Tests that protect an existing feature. A migration that used to take days and now takes hours. Documentation that would otherwise never get written. Trying out several approaches before committing to one.
That's real, and we use it. We pass part of that efficiency on to our clients: they get more software in the same time, or the same software sooner.
Where it leads you astray
AI doesn't make the decisions that matter. Whether a feature should be built at all. Whether a shortcut taken today will be expensive in two years. Whether a data model captures the exceptions that make this particular company what it is. Whether a solution opens a security hole that only shows up when it's too late. Those aren't questions of computing power. They're questions of experience and context.
Using AI without that judgment leads to a typical pattern: the first third of a project feels like magic, then it turns into a loop of debugging, structural problems and code nobody understands anymore. We regularly see projects that started as pure AI improvisation and ended up at a dead end. Fixing them then costs more than a clean start would have.
What this means for you as a client
The value of experienced developers doesn't drop in this world. It shifts: away from typing, toward deciding. That's exactly why senior experience matters more, not less. AI writes code faster than any person can. A person still has to take responsibility for it.
If you're commissioning a software project, ask the provider three questions. Who reviews the generated code before it goes into your system? Who makes the architecture decisions, and how are they documented? And how do the efficiency gains from AI reach you — in the price, in speed, or not at all?
AI-automatic means letting the machine run and hoping something usable comes out the other end. AI-native means using AI where it brings speed, and keeping judgment where it belongs.
