Introduction
Clara Reben reread the complaints after her start-up had already failed. That was the unpleasant part. The worse part was realizing that some customers had described the risk months earlier, only in language the team had treated as customer frustration.
Today Clara lives in Munich and advises young companies on AI governance. She knows the attraction of launching quickly; she also knows the cost of asking responsibility questions too late.
Story of the Path into AI
Her failed app had used a recommendation feature that seemed harmless in demos. In practice it made some options too visible and others almost disappear. There had been no proper opt-out, no clear explanation and no one person who owned the decision. After the insolvency Clara stopped telling herself that ethics had been a luxury they could not afford.
She worked through regulation, model cards, risk matrices and complaints handling. Lawyers spoke a different language from designers; developers wanted specific tickets, not principles. Clara’s first useful tool was an internal review protocol for small AI products. It asked about data origin, affected users, appeal routes and human supervision before the feature reached the launch meeting.
The first version was too abstract. A developer crossed out half the page and wrote: “What do I change tomorrow?” Clara kept that sentence.
Current Work
Clara now helps start-ups before risky AI functions enter the market. In one recruiting tool, she stopped a launch because historical data pushed certain applicants into weaker categories. The team replaced an automatic ranking with an assistant that gave reasons and required human review for borderline cases.
Her clients do not always enjoy the process. It slows decisions and makes uncertainty visible. But several have avoided reputational damage and, more importantly, can explain why a product behaves as it does. Clara treats documentation as part of the product, not as a folder for later.
Personal Advice
“Ethics is not decoration at the end; it is architecture at the beginning,” Clara says. Her advice to founders is blunt: if you cannot name who may be harmed, you have not understood what you are building.
Key Facts
Age and place: 31, Munich.
Background: failed start-up, insolvency, learning from product complaints.
Entry into AI: review protocol for small AI products.
Focus today: AI governance for start-ups.
Typical tools: risk matrices, model cards, audit workshops.
Werkstattnotiz
Clara keeps the crossed-out first protocol pinned above her desk. It reminds her that lofty language can hide from implementation. The line she still cannot automate is the one between a tolerable product risk and a business model that depends on users not noticing the risk.