AI in everyday work

AI that is still in use on Monday.

The usual AI course ends with understanding. Understanding lasts about ten days, after which the software is a chat window again. I build it the other way round: at the end of a day, something finished sits on your own device. A real task from your work, completed in full, in an environment you keep using without me.

Talk about a date +43 670 6549637

For businesses, for individuals and for training providers. Content and scope are agreed beforehand, the price is fixed before the date. The three routes

The sentence I hear most often

We all have one of those AI accounts by now. It has not taken any work off us yet.

That is rarely down to the people and almost never to the tool. It is down to the AI sitting in the browser while the work sits on the hard drive. As long as someone copies the question in and the answer back out, it stays a better reference book. It only gets interesting when the tool is allowed near your own files: reading, renaming, sorting, turning them into a table, writing a document. Every step visible, every step reversible. That is exactly the part I show.

What I offer

Three routes, depending on where you stand

The three hang together, but they start at different points. Anyone who has not begun at all starts with the workshop. Anyone already working and stuck usually needs a decision, not training. And anyone working with documents that must not leave the building ends up at the third question sooner or later.

01Workshops

Tailored workshops on AI, for businesses and for individuals. There is no standard slide deck I play back. I ask beforehand what you actually spend your day on, and build the day around that work. No prior knowledge needed, no programming skills.

Typical: A working day where AI stops being a chat window: first a guided task on prepared sample data, then the same pattern on your own real files, with individual support in the room. Plus the follow-on course for the one person in the business who will show the others afterwards. In-house for a whole team, or as an open date with a few places.

02Consulting on providers and ways of working

Which tool, which provider, and above all: for which task does it pay off and for which does it not. I sell no licences and earn nothing on a signature. That is why “you do not need AI for that” is an answer I can give.

Typical: Three subscriptions in the building and nobody knows which is for what. The question of which document an external service may even see, and how to turn that into a rule colleagues can apply too. A workflow built so that a mistake by the AI is noticed before it reaches the accounts. The assessment of whether a task is suited to AI at all.

03Local AI

AI that runs on a device at your place, without data leaving the building. I have been running this myself since 2024 and have tested what actually works usefully on existing hardware. That is the basis for an honest answer without a data centre, instead of reading out a brochure.

Typical: What runs on the hardware already standing there, and how much quality that costs against a large service. Which task stays local, which may go out. AI as a reliable step in an automated workflow instead of a gadget on the side.

How I teach

Four rules you can hold me to.

These are not advertising promises. They are the points where AI training usually fails. I have been running software rollouts for mixed groups since 2017, on site and remote, mostly with people who were not in the room by choice. That is where these rules come from.

At the end, something finished exists

Not a good intention, but a result: a working environment on your own device and at least one real task completed in full in the room. A day does not count for me if nobody has opened the tool again two weeks later.
Practice happens on your own documents

Sample data has exactly one job: to carry the first run while things can still go wrong. After that everyone works on real files of their own. What does not work in the room certainly will not work at home.
Limits belong in the course

I also show where AI fails, and how to build a workflow so that a mistake is noticed before it does damage. That includes the rule on which document an external service may see at all.
Nobody becomes a developer

Office workers should use the tool developers work with, because it is the only one that takes the step from suggestion to finished work. Nobody has to learn to program for that.

How a workshop runs

Three steps

Time in the room is regularly the smaller part of the work. Preparation for the specific business and following up afterwards are part of it, otherwise it stays a nice day without consequences.

Pre-conversation and setup

I ask beforehand what you work with, which files are involved and what is allowed on the devices. Plus a short remote setup session, so that nobody loses an hour installing things on the day itself.

The day

How the tool thinks and how to take it back. A guided task on prepared data. The same pattern on your own files with individual support. Filling your own scaffold. To close, the rule on which document may leave the building, and a demonstration of AI working entirely without internet on a device of its own.

Following up

A conversation around two weeks later. Not to ask whether you enjoyed it, but whether the setup has been used again since. That is the only number that really says anything about a course.

Where this comes from

What I have built with AI myself

I have worked with AI daily since 2024 and also run it on my own hardware. What I show in a workshop, I built myself first and keep using. Each of these projects can be demonstrated on screen in a conversation.

Local AI environment, since October 2024

An environment in which AI runs entirely on local hardware, with no connection to the outside. Tested what an existing graphics card can carry, set up an interface for everyday use, tried out adaptations. Built with Ollama, LM Studio, llama.cpp and Unsloth. That is where the solid statement comes from about what works usefully on which hardware and what does not.

Invoice reader for my own accounts, since July 2026

Reads incoming invoices, checks them against Austrian tax law, archives the originals and outputs the table the tax advisor works with. The design separates two things that usually get mixed up: the AI transcribes, the programme does the arithmetic. Anything that does not add up locks the document until a human has looked at it. Around 2,900 lines in 17 modules, 116 tests green, about 21 seconds per invoice, and no document leaves the machine.

Automated workflows with my own AI, since January 2025

An automation environment on my own server: messages are fetched, evaluated and sorted by my own AI, filed according to set logic, and depending on the result a notification goes out. AI as a reliable step in a workflow that is stable before and after it. Built with n8n and webhooks.

A knowledge base in which AI works without asking back

My own company documented so that a person or an AI can take it over: a glossary that fixes every central term, decision records with reasoning and rejected alternatives, standardised workflows and working instructions for AI agents. Around 170 files. An unfamiliar AI produces text in the right vocabulary from it, with no prior knowledge.

For training providers and institutes

Also as a trainer

I work with training providers and run courses on AI. I have been doing adult education since 2017, on site and online. Scope and format are agreed with the provider.

Enquire as a training provider

Where I say no

What I do not offer

No certificates and no qualifications. I am not an accredited training provider. Anyone who needs a certificate is in the right place at an institute, and I say so straight away.
No legal advice. Whether a particular processing is permitted is for lawyers to judge. I build the workflows so that the decision becomes possible at all, and say where I reach my limit.
No “AI will sort it”. For many tasks AI is the wrong tool, and for some it is simply too expensive. I say that beforehand, not after the day has been sold.
No building an AI from scratch. Setting up, adapting and comparing existing ones yes, developing a new one no. That needs different people and different data centres.

And if it turns out a workshop is not the right thing for you right now, I say so in the pre-conversation. Then it costs nothing.

Frequent questions

What people want to know beforehand

Do I need prior knowledge?

No. No prior knowledge, no programming skills. What you need is your own laptop that you are allowed to install something on, and the willingness to work with real files of your own.

Which tools are used?

The main tool is a development environment with an AI connection, which means an external service hosted in the United States. I name that openly instead of talking around it. The reason: AI running on an ordinary office laptop still works too unreliably once it is meant to touch files itself. That is why the rule on which file may go out at all is a fixed part of the day.

Does this work without the cloud?

For many tasks yes, and I demonstrate it too. For others it costs noticeable quality. Which ones those are, I tell you based on the hardware actually standing at your place, not based on a brochure.

What does it cost?

Workshops are charged as a flat fee, because preparation, the day itself and the follow-up belong together. I give the number after the pre-conversation, once it is clear what it is about, and before anything is fixed. Consulting on providers and on local AI is charged by time.

On site or online?

Both. On site in and around Vienna, online for everything that does not need hands on a device. For the first setup, one room with everyone in it has a clear advantage.

There are only two of us, is it worth it?

Yes. Small businesses often get the greater benefit, because nobody there is in a position to look into it full time. Individuals come to an open date, whole teams get one of their own.

Tell me what you are stuck on.

A few sentences are enough: what you work with, what you have already tried with AI and where it did not go any further. I come back with an assessment of which of the three routes fits and what it costs. I also say when none of them fits right now.