In operation since April 2026
What would a chief of staff take over first?
My chief of staff is an AI. It lives in my notebook, runs on my own server and has looked after taxes, mail, the rented flat and travel since April 2026.
Built. Running. Proven.
Counted when this page was built, as of 24 October 2026.
How it works
How does a notebook turn into a member of staff?
What it handles
Eight cases. Each with a number.
Three years of solar credits, overnight
Reminder out at 22:11, written commitment overnight at 08:29; roughly €1,100 paid out later.
Tax return, the AI as devil's advocate
356 items rated one by one, three genuine errors caught before filing, submitted five days ahead of the deadline.
Moving the server while it runs
35 services on one server with eight virtual guests; the document archive moved with roughly two minutes of downtime.
Inbox: 9,748 → 5,277
Inbox down from 9,748 to 5,277 e-mails; the target is permanently below 20.
Tenants' utility statement on request
2025 statement finalised on 7 Aug 2026, delivered, credit of €177.13 transferred to the tenants.
Paperless mail with a burden of proof
1,423 documents in the archive, paper backlog from 551 to 0, the archive's inbox from 466 to 0.
Owners' meeting prepared like a lawyer would
Discovered: 83 % of the 2027 service-charge increase is a heating advance, not the lift or the grounds.
Art on the wall that curates itself
31 works on the device; a new picture appears within ten minutes of the request.
How it came about
How does a director of software engineering end up building his own chief of staff?
My brother shows me what an agent does with a folder full of text files. Not chat. Work.
I create folders, write rules into them, hand the AI the key. The first request: a rental car for the summer.
If it works at home, it works at work. The experiment becomes an initiative with people who want to experience it themselves.
The AI advises on the purchase, sets up, backs up, monitors. Then the tax return, with the AI as challenger. Then three years of unpaid feed-in tariff: commitment overnight.
Every run makes the system sharper. Mistakes become rules. Numbers are counted; estimates carry a ≈.
Two years alone around the world, 200 conversations in 40 languages. The lesson from back then: context is what makes data valuable. True for travel. True for AI.
Context is everything
An AI is only as good as the context it is given. And that can be built.
Seeing is believing
Anyone who wants to lead AI has to use it first-hand. Only then is it clear what it should do for the team.
Knowledge grows when shared
The setup goes out as a template to family and colleagues, without a single private line.
At work
Leaders have to go first.
What of this transferred to a unit of around 300 people, and what did not. An observation, not an offer.
From transcript to ticket
Five to ten minutes after the meeting, minutes and tickets are on the board, no duplicates, correct parent ticket.
Critical risks current within a week
Around 100 active risks checked, around 90 tickets commented with proposals; within a week the owners had updated or closed a good half of them, and a good three quarters of the critical red risks.
Spec to running app in eight person-days
Eight functional and domain building blocks with around 160 input fields in about eight person-days; requirements become visible as running features immediately and can be reviewed by experts right away.
Legacy system as knowledge base: 250 requirements
Agents derive requirements from the legacy base and reconcile them with the team's current state; all content is draft until the expert review.
Honestly
And when it gets things wrong?
What comes next
What comes next?
- Every piece of correspondence becomes a caseWith a deadline, a follow-up and a ready-made reminder draft: the letter to the grid operator as the rule, not the exception.
- The family assistantSchool letters become appointments and packing lists. The meal plan fills the shopping basket. One page for my wife and me.
- The photo machineAfter a holiday, the 100 best pictures are on the phone the same evening. Curation is a metadata problem.
- Agents that run as a teamSeveral assistants in parallel, permanently in the background, taking orders where people already write: chat instead of tool.
- An operating model to pass onThe pattern can be handed over — the data stays with its owner.
- From experiment to unitWork is the harder test: approvals, confidentiality, change. How it goes on there is a conversation.
Built. Running. Proven.
Dr. Mathias Hentrich
Director of software engineering at a software company, unit of around 300 people. Before that strategy consulting, twelve years as an entrepreneur, two years alone around the world, a PhD in plant physiology.