AI with a human in the loop

We use AI where it makes us faster and fairer — and we deliberately leave it out where humans need to decide.

That distinction is not a disclaimer. It is the architecture. Every AI deployment we build sits inside one of four postures:

**AI does the work, human checks the result.** Routine tasks where the cost of an error is low and the volume is high. A model reads incoming documents and extracts structured data; a person spot-checks the output. Speed without blindness.

**AI sorts the evidence, human makes the decision.** Higher-stakes work where we need the model's range but not its authority. This is how we approach hiring: the model evaluates job-relevant competencies from the submitted materials — skills demonstrated, problems solved, experience applied — and calibrates its uncertainty explicitly. It surfaces the evidence. It does not rank candidates. It does not infer motivation, personality, or cultural fit. A human reads the evidence and decides. Protected attributes never enter the model's evaluation: that boundary is not a preference, it is a constraint built into the system.

**AI as thought partner, human holds the pen.** Creative and strategic work where generation is useful but synthesis must stay with the person who is accountable for the outcome. The model drafts; the human decides what is true.

**No AI.** Decisions that require a named human to own them — where accountability is not transferable and the value of the decision comes precisely from the fact that a person made it. Performance reviews. Redundancy decisions. Offers of employment. These stay human not because the model could not produce an output, but because an output without accountability is not a decision.

Jörg Tritthardt of Tritthardt + Richter (https://www.bfl-tr.com) — Ingenieurbüro für Leichtbau, roughly 40 years self-employed, world market leader in their field — put it plainly when we spoke: "KI ist ein Helfer und nie ein Ersatz für die geistig kreative Leistung — die Mischung macht es." AI is a helper, never a substitute for the creative intellectual act — the mix is what makes it work.

That is the posture we are after. Not "AI or human" as competing claims, but a clear allocation: here is what the model handles well, here is where the human is irreplaceable, and the boundary between them is made explicit rather than left to drift.

The four postures are not a philosophy. They are an operating discipline. What they prevent is the comfortable middle ground where AI does something consequential and no one is quite sure who decided it.