Key takeaways
- Technology is no longer the bottleneck: process, context and trust decide whether AI works in a company.
- The live experiment at the AI Superevent 2026 founded a company in three days but missed its 1,000-euro revenue goal. The reason was missing trust, not missing technology.
- Analyse the process first, then automate. Many workflows don't need AI at all, just a script.
- AI agents are only as good as their context: a central place of truth achieves more than a bigger model.
Three days, one venue in Vienna's Seestadt Aspern, around three dozen sessions: from 25 to 27 September 2026, the AI Training Institute's AI Superevent took place in Vienna. We at Degeler Consulting (DC) co-organize it, and Felix was on stage twice himself. Here is what we're taking home, boiled down to the points that matter for mid-sized companies starting Monday.
The shortest summary: technology is no longer the bottleneck. The bottlenecks are process, context and trust. Almost every talk ended up at one of those three points.
Founding a company in three days works. Selling doesn't.
The live experiment ran through all three days: a team led by Liam Schmid and Flo Huber founded a company on stage, with AI agents as staff. The goal: 1,000 euros in revenue by Sunday afternoon. Out of around 80 submitted ideas came an express website service for solo self-employed people, complete with brand, landing page, Stripe integration and ad campaigns.
The revenue goal was missed. According to the final update, the website had 800 to 900 visitors, the product was well received, and yet hardly anyone bought. The team's diagnosis: trust was missing. One detail says a lot: the legal notice page was among the most visited pages. People check who is behind an offer before they pay.
For us, that is the most important lesson of the weekend. AI shortens the path from idea to first campaign to under 48 hours. It does not shorten the path from first click to first purchase, because that path runs through trust. And trust isn't built with a prompt.
Understand the process first, then automate
Felix Schmidt showed, using a taxi company as an example, how daily trip planning shrank from two and a half hours to fifteen minutes. The first lever wasn't AI but process analysis: many workflows are deterministic and can be solved with scripts or low-code tools like n8n, Make or Zapier. AI comes in where it adds real value, for example with recurring dialysis or school transport runs.
Marcus Wagner reported on 7,000 recruiting calls made by an AI voicebot. His hard truth: technology is rarely the problem. Projects like this fail when the people on site aren't brought along. And the AI isn't a perfect machine, it's more like a colleague who makes mistakes.
That matches what we see in our projects: automate an inefficient process and you get a faster inefficient process. That is exactly where we start with companies.
Context is the real bottleneck for AI agents
On Sunday, two talks independently reached the same conclusion. Florian Schuenemann showed why AI agents make wrong decisions: not for lack of intelligence, but for lack of context, or outdated context. Overloaded context windows make agents less reliable, while a central place of truth with source, validity and timestamp makes them more reliable.
Vincent Mumme translated this to the personal level: a PersonalOS, meaning a structured folder with goals, projects and decisions, where every piece of information has exactly one home. An instruction file tells every agent how to work with the system, regardless of the tool. Automation only comes once that foundation is in place.
We have worked by this principle since long before agents were a topic. Why a single point of truth is the prerequisite for good collaboration applies to people and machines alike.
Faster isn't automatically better
A common thread that surprised us: several keynotes warned about the pace that AI brings. Esenija Schwerdfeger spoke about time abundance instead of time hunger. AI doesn't save time if we only speed processes up instead of rethinking them. Coordination and control then often cost more energy than before.
Marit Brademan introduced the concept of selective offloading: using AI deliberately to relieve thinking without replacing it. Offload without reflection and you risk an illusion of competence, meaning short-term wins with long-term loss of knowledge. Roger Basler de Roca closed Saturday with the productivity paradox: time saved almost always flows into more work, not into reflection.
Our takeaway: the decisive question isn't how much faster AI makes you, but what you do with the time you gain.
AI has a measurable footprint
On Friday, Felix and Phillipp Risch presented Graue Zellen: a tool that makes the energy consumption of AI use in companies visible. The core message: most people underestimate how much energy their daily work with AI costs, and providers publish hardly any transparent figures. That's why we measure ourselves and label every number as measured or estimated.
If you want to see what that looks like in practice: our own AI footprint is public and updated daily.
Privacy and security are solvable, fear of them isn't
Felix Schweinebraten put it in the title of his keynote: GDPR isn't slowing you down, your fear of it is. AI agents should be treated like employees, with clear permissions, controls and human review of results. Liability for wrong output lies with whoever uses the tool, not with the provider.
Falko Werner demonstrated live that applications built by vibe coding need the same security checks as classic software: no API keys in the code, security tests before and after going live, protection against prompt injection. Adrian Schmidt and Stefan Starflinger showed how a business can run on local open-source AI, so sensitive data never leaves the building in the first place.
Websites are now built for people and agents
The talk show on Friday afternoon with Gaston Geilenkothen, Markus Hartlieb, Raphael Schaad and Felix asked where mid-sized companies are heading. One thesis stuck: traffic from AI agents has already overtaken traffic from humans. Websites therefore have to work for both audiences. Fittingly, the book Gebt der KI ein Zuhause was presented and handed out to all attendees.
Ramtin Ghasemipour-Yazdi took the idea further on Sunday under the heading Zero UI: agents select products, check budgets and make purchases without any classic user interface. SEO shifts from keywords for people to structured data for machines. That is exactly why we build articles like this one with clear answers, FAQ and schema markup.
The most important things happen between sessions
As packed as the programme was, what sticks are the encounters. At Friday's blind dinner, an AI put together tables of six in Viennese restaurants, and you only found out who you were eating with on arrival. On the side stage, the Young Explorers aged 15 to 21 had their own programme. Add a poetry slam about AI and bureaucracy, and the Community Awards for the people who pull others along all year.
The AI Superevent is set to return next year. If you want to be there, you'll find everything at the AI Training Institute.
