Praxis
The engagement: built, handed over, running
Your team runs real work through AI agents, with a human gate at exactly the points that matter, and you can prove what the agents did.
Not “we use AI”. Not a chatbot. Work that agents pick up, do and finish, where a person approves the plan once instead of approving every step, and where nothing irreversible happens without a named human.
I do not build it and disappear. Before I leave, your team learns to run it, change it, and notice when it is wrong.
Scoped to your operation
3-month minimum
I do not publish a number for this. What it costs depends on how much of your work is going through agents, how many points need your yes, and what your current tools already do. I look at the actual case first, then quote it.
Five things, and each is a thing rather than a document.
- 01
A map of where it stops and asks
Every action your business takes, sorted into two piles. Reversible, which an agent does without asking. Irreversible, which waits for one named person to say yes at the moment it becomes irreversible: money, data, passwords and logins, anything public, anything deleted. This is the piece that makes speed safe, and almost nobody has it written down.
- 02
A board that updates itself
Work lives in one place, and its status is read from wherever your work really lives: your inbox, your spreadsheets, your project tool. Reaching a conclusion and recording it become the same action. Nobody updates a tracker by hand.
- 03
Agents with named jobs, and a clear way to give them work
Not one do-everything chatbot. Each agent has a named job and clear limits, works on its own piece without touching the rest, and hands back something you can check. Plus the rule that keeps it safe: work nobody planned cannot give itself permission to start.
- 04
Early warnings
Automatic checks that warn you before something quietly breaks, while it still costs nothing. A login that expired without telling anyone, a service you rely on that was switched off, a setting that no longer matches how things actually run. Each of these exists because it caught something real here.
- 05
Spending limits, a record of every action, and an off switch
A spending limit that really stops an agent when nobody is watching. Every action traced back to the decision that allowed it. One switch that stops everything.
The ninety days
Month one
Map and agree
Your work mapped. The rule for where it stops and asks, written and agreed. The board set up and reading wherever your work really lives.
Month two
First agents at work
The first agents doing real work in your business. The automatic checks switched on. Your team running it day to day, with me alongside.
Month three
Hand over
Spending limits and the off switch live. Your people running it without me. Written down well enough to survive my leaving.
What it is not
- Not a chatbot set-up, an AI tool rolled out to everyone, or prompt training.
- Not staff augmentation. I am not doing your work; I am building the thing that does it.
- Not a software licence. You own what is built.
- Not a pilot. Three months to something running for real, or it has not worked.
What I need from you
- One decision-maker
- Someone who can agree the rule for where it stops and asks. Without that, the engagement produces a document instead of an operation.
- One real piece of work
- Not a trial run. A map of where it stops is only useful against work that has consequences.
- Access to where your work really lives
- Your inbox, your spreadsheets, your project tool: wherever the real record is kept.
Nothing is copied to my machine. My access ends at handover.
What it is worth
- Against a hire
- A senior AI engineer in Western Europe costs €85,000 to €140,000 in base salary alone, before employer contributions or recruitment, and needs months before they are useful. The engagement has your operation running by the end of month three.
- Against a consultancy
- A three-month transformation engagement ends with a deck and a pilot. This ends with a running operation your people own, still running after the engagement closes.
- Against one unchecked action
- A deleted customer list. A wrong payment. Something published that should not have been. Each one is a cost you cannot take back, and each one is exactly where it stops and asks.
One senior AI engineer, base salary only. By month 3 a US hire has cost €57k and a Western European hire €28k. The engagement ends there. The hire does not, and by month twelve the same person has cost €228k and €113k.
This understates the alternative on purpose. It excludes employer contributions, recruitment, equity, management time, and the months spent hiring before anyone starts. Every one of those makes a hire cost more.
United States: $220,000–$310,000 base · €189k–€267k · Senior AI engineer base range, 2026 market surveys
Western Europe: €85,000–€140,000 base · Senior AI engineer base range, Germany, 2026 market surveys
Converted at 0.8604 USD to EUR, 8 September 2026.
I have run my own company this way for months. A team of named agents, one yes on the plan instead of a yes on every step, a board that updates itself, automatic checks that have caught genuinely expensive things, and a hard rule that agents stop before anything that cannot be undone. The proof is not a case study. It is that the operation exists and I can show it working.
After the three months
Maintenance, if you want it.
Scoped with you when the three months end, and written into your contract rather than onto this page. What it costs depends on what was built and how much of it is still moving.
Available to engagement clients only. It cannot be bought on its own.
What it covers
- Keeping the automatic checks working when a service you rely on changes or shuts down
- Moving where it stops and asks when your business changes shape
- Adjusting what each agent does
- Working out what happened when something stops behaving as it should
What it does not
- New capability. A new area of work, a new agent, or connecting a new tool is a new engagement, not maintenance.
How can I help?
I'm Ludmilla. I read every one myself.