AI consulting for the Mittelstand
I have been writing software since 2000 and work with AI agents every day. I show your team how to get started in practice and how to measure and verify the results.
The starting point
AI experiments are running in many companies. The results are sometimes impressive and sometimes wrong, and nobody can say which is which. So pilots stay pilots and never reach daily operations. What is missing is a method that builds trust in AI step by step, backed by measurements.
Approach
Every rung up hands the AI more responsibility. Trust is earned with evidence.
Based on Matt Pocock's Trust Ladder, adapted from my daily practice for companies. Matt Pocock at aihero.dev
Rung 1
Give Claude Code almost nothing except a well-known problem. The result surprises: it works, it is structured, and it comes with docs and tests you never asked for. That result becomes your baseline.
Prompts such as "Build Snake as a browser game" or "Build a ticket system like Jira, with tickets, sprints and a sprint board."
"Create an onboarding checklist for new employees."
Rung 2
When results in real projects fall short of the baseline, ask what is holding the AI back here. Fix the harness, not the model: context, docs, conventions, a second brain.
Quality drops in an existing codebase. Add context, docs close to the code and conventions.
Answers to customer enquiries sound generic. Put product knowledge, tone and examples into a second brain.
Rung 3
Make trust measurable: a gold standard of real cases, plus tools that make checking AI output fast.
A gold standard of real cases, checked automatically for correct, wrong and missing values.
A review tool shows every extracted invoice line next to its source. Approval takes seconds.
Rung 4
AI takes over real workflows. Rungs 3 and 4 blend, because every automation brings its own review check.
Agents handle recurring tickets, and every run has its own review check.
Incoming mail is classified and prepared, and a human approves.
Services
Four offers, each tied to a rung of the ladder. We begin where you are.
Rung 1–2
In hands-on workshops, your team reaches the baseline from rung 1 within days. That includes clear rules for safe use: which data goes where? The basis is a 12-module Claude Code curriculum. It also works for people who do not write code.
Rung 2
We find out what holds the AI back in your real work. Then we build the context it needs. "20 % tool, 80 % habit": the habits in your team matter more than the software.
Rung 3
Real cases become a gold standard. With it, you measure correct, wrong and missing values separately. Review tools let your people check AI output in seconds.
Rung 4
I automate selected workflows, each with its own review check. Alternatively, I take the role of your external, part-time AI lead. I do this alongside my main job, so my availability is limited.
I deliberately take on only a few engagements at a time.
Proof from practice
There are no client logos here. This is my own work, and you can trace all of it.
Own product
An independent MTB trail platform for the DACH region. I build it solo in my spare time with agent-based workflows (Nuxt, Supabase, PWA).
516 spots: 307 trails, 97 bike parks, 112 dirt parks
Measure, don't guess
For the AI data extraction in Trailradar, I measured a model on 286 hand-rated trails, tracking correct, wrong and missing values separately. A cheap model reached 84.8 % accuracy but missed 42 % of values. I rejected it. No data beats wrong data.
84.8 % accuracy, with 42 % of values missing
Hackathon
On 5 October 2026, a team of five built Briefklar in the "AI for Good" track: an explainer for German official letters with on-device OCR, so the data stays private.
5 h from idea to Briefklar
Talk
I gave this talk in-house to a development organisation. It covers how teams work with AI agents and how to hand them responsibility in sensible steps.
Training
A curriculum that runs from installation to agent workflows. It is the basis of my workshops.
12 modules, from installation to agent workflows
About
Senior Full-Stack Software Developer, driving the AI transformation
I have been developing software since 2000, many of those years in industrial companies. Today I work at a corporate startup in industrial software.
I learn how AI agents work in practice by building with them every day. In my free time I ride mountain bikes, and that is where Trailradar comes from.
I run the consulting alongside my main job. That is why I deliberately take on only a few engagements at a time.
Contact
Tell me where you stand today. I reply personally.
Thank you! I will get back to you within two working days.