AI that stays when
the consultant leaves.

For getting started, the free versions of Claude and Gemini are enough, 0 EUR combined. More important than the tool is the rule: one tool, one task, one week. After that you know more about your needs than any comparison list could tell you.

That depends on your daily work, not on benchmarks. If you live in Google Workspace, Gemini is the shortest path. For German texts and document work, Claude is strong. ChatGPT remains sensible if your environment already works with it and you want to keep using what exists.

Between 0 and 25 EUR per month. The free versions are enough for testing, paid plans start around 5 EUR and the big assistants sit around 20 EUR per month. It only gets more expensive once you need automation and team features, and by then the benefit should be long proven.

No. “I can't code” has become the most expensive excuse you can afford. You describe to the AI in plain language what you need, it writes the code and corrects itself. The real skill today is: being able to describe clearly what you want. And you are often better at that than a developer.

With a single question: which task eats most of your week? You take exactly one of those and set it up in a single afternoon. The tools come after that. One working workflow motivates more than ten abandoned experiments.

Yes. With tools like LM Studio or Ollama, a language model runs directly on your machine: data never leaves the house, running costs are zero. The results are weaker than the big cloud models, but for sensitive content local is often the right choice anyway.

With five questions instead of a tool list: What costs the most time? What depends so much on one person that everything stops without them? Where has your gut been saying “this should be easier” for years? What would your team automate first? Which data sits around unused? The answers that overlap are your starting point.

A prompt tells the AI once what to do. A skill tells it permanently how to think and work: your tone, your rules, your workflow, stored once instead of briefed every time. That is exactly where the difference between “trying AI” and “AI works for you” begins.

With a fixed instruction in your project. Mine reads: “You are my critical sparring partner, not a cheerleader. Start every assessment with the three biggest weaknesses or risks, only then the strengths. Praise only when you can tie it to a concrete result. If I am wrong, say it directly and explain why.” The difference is enormous: “Great idea!” becomes “This only pays off from 40 participants.” That is something you can work with.

Agents are AI that works through tasks on its own instead of just answering questions: you brief them, let them run, and review the result. For the self-employed this means delegating without hiring. But it is not autopilot. I review every result before it goes to a client. Human in the loop is non-negotiable.

Yes, that is happening right now, and that is exactly your chance. Use AI for research, rough drafts and structure, the first 50 to 70 percent. Your voice, your judgment and your examples do the rest. The better AI gets, the more valuable the part that is recognizably human becomes.

In the free version: no. The rule of thumb is “the case, not the person”: describe the situation, not the human behind it. For company data you need at least a paid plan, a data processing agreement with the provider, and AI training switched off.

No. Free versions often use your input for training and offer no data processing agreement. The minimum for business use: paid plan, data processing agreement, training opt-out. With the big providers that is set up in a few minutes.

Not in general. Under the EU AI Act, a visible labeling duty has applied since August 2026 mainly to deepfakes, meaning deceptively real depictions of real people or places, and to unreviewed AI texts on matters of public interest. If you review texts editorially and take responsibility for them, you usually fall under the exception.

Less than the headlines suggest. Most small companies are deployers, not providers of AI systems. What matters most: AI literacy in the team, the transparency duties in force since August 2026, and a quick look at the risk class of the systems you use.

Yes, on several levels: EU providers like Mistral with Le Chat, platforms with EU hosting like Langdock, or fully local models on your own machine. Often, though, the US provider with the right settings and contracts is fine too. What fits depends on the kind of data you handle.

Between 0 and 3,890 EUR net, depending on the level: the KIP Tool-Kompass is free, the KIP AI Recipe Database costs 119 EUR, the KIP AI Workshop 497 EUR, Workshop PLUS 1,890 EUR and Sparring MAX 3,890 EUR. All flat prices: no subscription, no hourly rate.

Because otherwise you get a system you do not understand and a service provider you can never get rid of. If we build together, in the end both belong to you: the workflow and the ability to build the next one yourself. That is the whole point.

No, but commitment. You should be ready to type yourself and to complete small tasks between sessions. If you only want to watch, I am the wrong person, and I will say so in the compass call just like that.

30 minutes, remote. Beforehand you answer five short questions so we do not start from zero. The call is about your situation and your most expensive time sink. Afterwards you get a clear recommendation, even if it reads: you do not need me.

Yes. Not convinced after the first 30 days that we are on the right track? You get your money back in full. No ifs, no buts.

Remote is the default, and it works excellently for sparring and workshops. For team workshops I also come on site, anywhere in the DACH region, with travel included in the flat price. Group formats run in German; one-on-one formats and team work are available in English as well.