Background · Founder
Reto Lutz.
A single point of contact: hands-on AI training for companies in Switzerland, directly with the founder.
Why ai-edu.ch?
My background: IT project leadership in a highly regulated Swiss industry - an environment where data protection, auditability and clean system handovers are mandatory. Today I work at the intersection of engineering and live operations - where a solution proves whether it holds up in daily work or only shines in a workshop.
In parallel I study Business AI at FHNW and deliberately run ai-edu.ch as a one-person operation. No agency overhead, no account manager between you and the person who delivers the training.
Basis for trust
Honest, before there are references.
When a provider is new, the answer is not to claim louder. It is to show more clearly how the work is done.
- 01 No fake references
- ai-edu is in an open pilot phase. That is why you will not find invented client logos, generic testimonials or anonymous success stories here.
- 02 Proof through a demo
- Instead, you see before you book how the training works on realistic Swiss demo data.
- 03 In writing before you commit
- Before a workshop, you get a concrete curriculum with format, roles, data frame, exercises and expected outcomes.
The leap is not in intelligence.
What a system achieves depends less on how clever it is - and more on what it is allowed to reach.
Most of the attention goes to model generations and benchmarks. I hold that to be the less interesting axis. What convinced me was not a cleverer model, but the point at which one stopped merely answering: it takes on a task and works through it over many steps, on real files, with real tools. That is where my training puts its weight - not on the single prompt, but on the question of which work a system may take over, and under which rules.
That shifts the security question. A chat window where someone types a sentence is a manageable risk. A system that acts on your behalf reaches everything the account it runs under reaches - including the shared folder nobody has thought about in years. Permissions are the agent's radius. To me that is not a compliance exercise at the end, but the first question before anything goes into production.
How seriously I take that is something you can read rather than take on trust: for Claude Code I put together what the vendor documents actually say about company data - with a chapter of its own on the three questions I cannot answer. Where a source is missing, it says so, instead of a reassuring phrase filling the gap.
In the training that means: exercises run on invented company data, never on yours. I built those data sets myself - the check digits of IBAN and UID compute correctly, the numbers behind them are made up. Realistic enough for the exercise to hold, without a single real personal record having to leave the room.
Stance
No. 03 · ai-edu
What I stand for.
Four points that frame every training and every engagement as a working rule.
- 01 Practice-first
- Every session works on real tasks from your Swiss business day - real in the brief, fictional in the data sets.
- 02 Regulation-fit
- DSG and data residency: local requirements frame every workshop. I go through industry-specific obligations in the intro call.
- 03 Partner, not vendor
- After the workshop I stay reachable by email. Plus the tool radar: personally curated pointers to tools that fit your stack, no generic newsletter. Scope depends on the format.
- 04 Demonstrably current
- The AI landscape shifts weekly. I keep the content current - from my studies and from live operations.
- Daily
- Claude and Claude Code. This website, its articles and my training materials are built with them - you can watch that in the showcase, which holds real, shortened recordings rather than re-enactments.
- Regularly
- ChatGPT, Microsoft 365 Copilot, Gemini, Perplexity and GitHub Copilot - including their command-line versions where those exist. That is precisely the point for a training session: anyone who only knows a tool through the browser misses half of what it does, and undoes, inside a company.
- Examined, not recommended
- The cloud platforms behind them - Azure OpenAI and Vertex AI - I look at, because the question of Swiss processing always comes up in regulated companies. I do not run them for clients and I do not set up tenants.
The question that gets asked before a Copilot rollout.
Microsoft 365 Copilot makes visible what has been set wrong in permissions over the years. A payroll table shared with "everyone in the organisation" at some point used to be practically unfindable. Afterwards, one question in the chat is enough.
That is why this question comes before the licence question - and why the answer to it is not a footnote. I still do not audit your permissions. That is work for your IT or their partner and does not fit into a training day. What I deliver is the list of questions that need answering before a release.
Anthropic Academy
- AI Fluency series: Framework & Foundations and Teaching AI Fluency, plus the editions for educators, students, pK-12 (incl. Train the Trainer) and Small Businesses
- Claude Platform 101
- Introduction to Claude Cowork
- Building with the Claude API
- Claude Code 101
- Claude Code in Action
- Introduction to Model Context Protocol
- Model Context Protocol: Advanced Topics
- Introduction to Agent Skills
OpenAI Academy
- AI Foundations
- Applied AI Foundations
- Agents and Workflows
Google Cloud Skills Boost
- Introduction to Generative AI
- Introduction to Large Language Models
- Introduction to Responsible AI
Next step
Ready for an intro call?
Thirty minutes, a first assessment: where AI holds up in your operations and where it does not.
Free · 30 min · no obligation · no sales script