Keynote · AI agents · 30–60 minutes · EN · DE · FR
AI agents: when software starts to act.
A talk for organisations deciding in 2026 which work to hand to agents. From someone who builds agent systems for hospitals, where an unnoticed error is expensive.
Until recently AI answered. Now it does: it researches, writes documents, coordinates dates, triggers the next process step and calls other programs on the way. That is the difference between an assistant and an agent, and it changes who in the organisation is responsible for what.
My company builds systems in which several agents check each other’s results before anything reaches a patient record. From that work I know what agents reliably do today, where they regularly fail, and which kind of error is the more dangerous one. Everyone talks about hallucination: the machine adds something that was not there. Omission is more common: the machine leaves something out, and nobody can notice what is not on the page.
The talk gives the room a map of which tasks agents take over today, which in twelve months and which not for a long time. It shows the three controls without which no agent belongs near customer data, contracts or payments, and it answers the question every room asks: how much does the AI still make up, and what do you do about it?
What the room leaves with
A map, not a forecast
Which tasks agents take over today, which in twelve months, which not for a long time, sorted by what can go wrong.
The three controls
Clearance of the data, checking of the results, limits on the acting. Without these three no agent belongs near real data. With them, getting started is a decision of weeks.
Who answers for what the software decides
Whoever signs, checks. How organisations settle responsibility once an agent has triggered the next step, and which part of that belongs in the AI policy.
Which events this is for
- IT and digitalisation conferences
- Leadership meetings and board days
- Association and chamber conventions
- Kick-off of an AI or automation programme
- Mid-sized companies and corporates, all industries
Formats
Keynote
30, 45 or 60 minutes, with or without Q&A. Examples from your industry, agreed in the scoping call.
Keynote + working session
The talk in the morning, ninety minutes with your leadership in the afternoon: which five processes go to agents first, and with which controls?
Leadership workshop
Half or full day. Your people build, on one of their own cases, the controls an agent needs in your organisation, and leave with a clearance list.
Why me
- I build agent systems for hospitals, where a wrong output is a clinical and a legal event.
- In my own company I let two senior experts go and took over their jobs with AI. I know the tools from daily use, with numbers.
- My company is preparing the rollout of clinical AI across about 50 hospitals. The controls that hold there hold in your organisation.
Sven really brings fresh solutions to old problems. You can see he is pushing AI to its edges for very pragmatic problem-solving.
An intelligent mix of practical examples and scientifically grounded knowledge — refreshing.SVP Roche Diagnostics Germany
Questions organisers ask about this talk
Does the audience need technical background?
No. The talk explains through everyday examples what an agent does and does not do. People who build software still take something away: the controls are rarely sorted cleanly even among specialists.
Do you show tools live?
Examples yes, product pitches no. On request I show an agent at work on an anonymised case, if the venue’s connection carries it. Aiomics is disclosed as an interest and otherwise not mentioned.
How current is the talk?
Agents change monthly, so the talk is updated before every date and tuned to your industry in the scoping call. Slides and abstract reach you a week ahead.
Which work do you hand to agents first?
Send the date, the audience and the processes in question. Athenas manages my calendar and answers inside the working day.
