Five weeks have passed since the second edition of AI Expo Europe, held on 1 and 2 November in Bucharest, at the Radisson Blu. I let some time go by on purpose. In the heat of the moment, any event looks like either a triumph or a disaster, depending on how much sleep you got the night before. Now I can look at it more calmly.
Last December I wrote looking back on our first year of AI Expo Europe. That piece was about building something from scratch. This one is different: it is about what changes when you reach a second edition, and about how the conversation around AI has shifted in the meantime. Because it has shifted a lot. Once again we had dozens of speakers and exhibitors from Romania and abroad, and what I heard on stage and in the hallways confirmed a few things and surprised me with others.
From “what is AI” to “how do we put it into production”
The first difference was visible within hours. In 2024, many conversations started with basic questions: what it can do, what it cannot do, whether it is a passing fad. Fair questions, and part of our job was to answer them.
In 2025, almost nobody asked “what is it” anymore. The questions were “how do we implement it”, “how do we measure whether it works” and “who is accountable when it gets things wrong”. In other words: implementation, measurement and governance. To me, that is the clearest sign the market has matured. When people start asking about processes and responsibilities, they are no longer talking about a toy. They are talking about a work tool.
Agents and workflow automation: the theme of the year
If I had to sum up this edition in one word, it would be “agents”. Not chatbots that answer questions, but systems that see a task through in several steps: read an email, check stock, prepare a quote, send it for approval. A year and a half ago I wrote that AI agents are the next step after chatbots, and this year that step became the main topic.
What I liked is that the enthusiasm came with a healthy dose of caution. I heard the same common-sense rules more often than I expected: start with internal tasks, grant limited permissions, keep a human who signs off on anything irreversible. It is exactly what I described in my AI automation case study for a small company. The gap between an impressive demo and a system that runs for months without surprises is, in fact, the boring work in the middle.
What companies are actually asking
I spent a lot of time in the hallways, just listening. The questions kept coming back, almost in the same order:
- ROI. “How much does it cost and when do I get my money back?” It is the right question, and those who skip it risk running vanity projects.
- Data quality. Many companies discover that the problem is not the model but the data: scattered across ten files, incomplete, contradictory. AI on top of bad data produces fast, convincing mistakes.
- Security. What data leaves the company, where it ends up, who has access. A question that rarely came up last year and is now among the first.
- The AI Act. Since August 2025, the obligations for general-purpose AI models have applied, and that made the EU regulation feel a lot less distant. When it was adopted I wrote about what every company should know about the AI Act, and today’s questions are far more practical: “what do we need to do, as a company that only uses these tools?”
The bottleneck: people, not technology
If I had to name a single bottleneck, it would not be technology. The tools are available, many are cheap, some are even free. The bottleneck is people who know how to use them well: who understand the company’s processes, what a model can do, and where it has to stop.
I felt this in almost every conversation with managers. They are not necessarily looking for “an AI specialist”. They are looking for people on their own team who can learn to work differently. That is why I believe ongoing internal training, built on the company’s own examples, will matter more than any subscription to a new tool.
The region has real use cases
One thing I am genuinely glad about: companies and startups from Romania and the wider region no longer come just to listen. They come with projects already running, with real difficulties and with solutions found on small budgets. Last year I wished for more case studies from the region. This year I felt they exist; they just need to be brought into the spotlight.
The role of the public sector was on the table again, too. The state is at the same time the one writing the rules, a large customer and a potential user of AI in public services. I think direct dialogue between those who build products and those who make the rules is one of the most useful things an event like this can offer.
What I learned as an organiser in year two
A second edition teaches you things the first one cannot. I am writing them down for myself, but they might help anyone building a community:
- Consistency matters more than size. The first edition proves you can do it. The second proves it was not a fluke. For speakers, partners and attendees, the fact that you come back is a stronger signal than any number.
- Practical sessions carry the most value. People go home with what they could apply, not with what they admired. That remains our direction.
- Networking is not the break between sessions; it is part of the programme. The best conversations happen in the hallway, and a good agenda leaves room for them.
- Community is built between editions. A two-day event once a year is not enough. What you do in the other 363 days decides whether people come back.
What comes next
AI Expo Europe continues in 2026. I am not giving a date yet, for the same reason as last year: I prefer to announce things that are certain. What I already know is the direction: less “what is possible” and more “what works”, with the focus on implementation, on people and on the rules we all have to live with.
Looking back at what I wrote a year ago, I am glad I was right about one important thing: business people want practical AI, not philosophy. In 2025 they came with even sharper questions. It is our job to come back next year with answers to match.
What AI question would you most like to see answered at a future edition: one about money, data, people or rules?