In a nearly two-hour episode of the Finnish investment podcast Sijoituskästi, Tero Ojanperä — co-founder of Silo AI and former CTO of Nokia — sits down to explain where artificial intelligence actually stands right now. The episode’s core message is easy to summarise: if AI is still just a chatbot to you that you type questions into and read text answers from, **you are about two years behind**. The shift has already moved on to the next phase, and that next phase is exactly where Finnish founders either consolidate their position or end up watching from the sidelines.
Sijoituskästi is one of Finland’s most-listened business and investment podcasts, and the choice to invite Ojanperä as a guest tells you the topic now matters to people who do not follow developer forums. Good — because the content of this episode is not aimed at engineers. It is aimed at decision-makers.
Source: Sijoituskästi — What you need to know about AI ft. Tero Ojanperä, 22 May 2026 (in Finnish)
What AI actually is — and why it hallucinates
The opening of the episode is useful for everyone, because Ojanperä takes the time to lay out the foundation that few really master: what current AI actually is. A language model — the engine behind ChatGPT, Gemini or Claude — is not a database, and it is not a search engine. It is a statistical model trained on vast amounts of text to predict which word is most likely to follow the previous one. When it answers your question, it does not “look up” the answer. It constructs the answer one word at a time, based on what it has learned about how text is structured.
This is also the reason AI sometimes hallucinates — calmly inventing facts that simply are not true. The model does not know what it does not know. It just predicts, word by word, what sounds plausible. If you ask the revenue of a small company the model has no data on, the answer will look credible — because the language model knows how to construct credible-looking sentences — but the figure may be completely made up.
This is the point at which many founders have already slipped up. There is no need to fear AI. There is a need to know what it is good at and where it is not. Ojanperä explains this with a precision that works for non-specialists — and that is what makes the episode worth your time. The limits of AI are not going away in the next few years. They are merely moving a little further out.
Agents — the next wave, already underway
When the conversation moves past chatbots, the pace picks up. AI agents — systems built on top of AI that take actions on behalf of the user, rather than only answering questions — are, in Ojanperä’s view, the part of the field that now deserves the most attention.
A chatbot answers a question. An agent handles the task. For example: a chatbot tells you how to book a ticket. An agent books the ticket, compares prices, manages your calendar and sends the confirmation — all from the one sentence you wrote. Especially in business use, this moves AI to a completely new level. Customer service, sales, procurement, bookkeeping, reporting and marketing tasks become automated wholes that don’t just write text — they decide, act and report.
The episode’s message is that agents are no longer experimental. They are being built into production right now. For a Finnish founder, this means two things. First: if you have processes built from repetitive, rule-based steps, they can be automated now. Not in two years. Now. And second: if your competitor does this before you, the lead is longer than you might think, because agents learn from the tasks they perform.
Silo AI — when a Finnish company rose to Europe’s top tier
The episode’s most personal section comes when Ojanperä tells the Silo AI story. He founded the company in 2017 together with Peter Sarlin in Helsinki. The timing was off. AI was academically interesting but, for businesses, still distant — and certainly distant from companies that buy software on annual budgets rather than as research projects. Silo AI still set out to build AI for businesses, not for consumers.
Years of work paid off in a way few people dare to imagine. American chip giant AMD acquired Silo AI in the summer of 2024 for roughly 665 million dollars. It is one of the most significant deep-tech exits in Finnish history. Silo AI was built in a European way, Helsinki-first, and for the long haul — eight years of work before the spotlight arrived.
To be honest about it: Silo AI is not a one-size-fits-all model for every Finnish startup. The company deliberately chose the harder path — selling enterprises something they did not yet know they needed. For a Finnish company this path requires patience, a long runway and world-class experts. But it is a possible path. It is a proven path. And it is a path built from Finland.
What bottlenecks remain
Toward the end of the episode, Ojanperä addresses the place where the conversation usually drifts: where the real limits of AI actually show up. The worry he voices in the episode is not that AI is running out of capability. The worry is elsewhere.
Energy. Large language models consume electricity in a way that reshapes the energy markets of entire regions. In the US, the hyperscalers — the data-centre operators in the AWS, Microsoft and Google league — are signing their own nuclear deals. In Europe and Finland this is both a risk and an opportunity: we have affordable, low-emission electricity, but we also have limited capacity.
**Data.** The best language models have already been trained on nearly all of the high-quality text available on the internet. The next generation needs new kinds of data — and this is precisely why business data, until now considered “extra”, is becoming extremely valuable. The next-phase winners of AI will be the companies sitting on the best data in their industry. A Finnish company should now ask itself what its data really is and who owns it.
Trust. Hallucinations are not going away. Instead, we are building systems in which AI is checked by AI, and in which a human is always in the loop at a key step. This is slow, expensive and necessary — especially in regulated industries.
What to take away
Three thoughts from the episode are directly useful for a Finnish founder:
- Don’t get stuck in the chatbot phase. AI is not a tool for writing marketing copy faster. It is a way to rebuild whole workflows. Ask which of your processes could be done by agents rather than people — and start a pilot now, not next year.
- Think about what your own data is. If you are building a B2B company, the specialised data that builds up with customers will, in the next few years, be a central part of your competitive moat. Don’t sell it off, don’t mess it up, don’t let it leak. Treat it like a bank vault.
- **The long game pays.** Silo AI was not built by chasing a two-year trend. It was built by eight years of work in enterprise service. Finnish growth-entrepreneurship culture has something that suits this game: persistence and engineering depth. Use them.
This Sijoituskästi episode is unusually useful for what it does not do. It does not sell AI hype. It does not promise that every company can 10× its productivity in a month. It explains where AI actually is, why it hallucinates, what it is good for, what it is not yet good for — and how Finland can play its role well from here on.
If you are a founder, an investor or a member of an executive team, spend the two hours. You do not need background. You only need what Ojanperä is trying to convey throughout: curiosity to look a little further than the headline.
Innohub TV
Watch this content also on Innohub TV
We picked clips from Innohub TV that continue the same topic.
