Codento’s Self-Driving Enterprise together with Google Cloud featured in today’s Kauppalehti – a leading Finnish financial magazine.
Here you can read the English translation of the original Finnish article:
Finnish companies are making one mistake with AI – “We’re thinking too small”
Markus Hongisto, Country Manager at Google Cloud, and Anthony Gyursanszky, CEO of Codento, outlined what an organisational agent team looks like and explained how companies should keep pace with developments in artificial intelligence.
Anthony Gyursanszky, CEO of IT consulting company Codento, enters a task to create an agenda for an upcoming meeting in an imaginary scenario for a CEO and their executive team working at a company listed on the Nasdaq Helsinki. The AI agents perform their work in real time, and the topics to be addressed for the meeting appear on the screen in order of priority. It takes two to three minutes for the agenda to complete. The agents know how to take into account constraints essential to the company—for example, in a defense company, regulations from NATO and the target country.
In the view of the Self Driving Enterprise Blueprint (SDE Blueprint), you can check which agents out of dozens of different problem-solving tools were used. It is precisely agentic AI and self-improving loops that are currently all the rage in AI development. This enables work to happen in the background without requiring virtually any human intervention. Furthermore, in a properly built system, the time previously spent on prompting—which often interrupted work—is reduced, as agents only need to be fed an idea of what problem to start solving.
Codento’s SDE Blueprint is built on the Gemini Enterprise platform. Markus Hongisto, Country Manager for Google Cloud, estimates that AI development currently takes place primarily in cloud services, which is where the latest models are built. ”Cloud services are clearly leading the direction in AI development. With capabilities advancing at a breakneck pace, keeping AI development in the cloud enables customers to benefit from the platform’s continuous evolution. Bringing similar capabilities to traditional environments is worth considering only in very specific cases.”
According to Hongisto, cloud environments enable multiple models to be used simultaneously for the purposes best suited to them. Hongisto notes that although Gemini Enterprise operates natively with Google’s own Gemini models, the Gemini Enterprise Agent Platform gives customers the flexibility to also use language models from other providers to boost specific agents and tasks.
For example, standard chat features or light coding do not necessarily require the most advanced Anthropic or OpenAI model, even if it has been proven most effective for the most demanding tasks in testing. ”I’ve used the example that Wolt couriers use electric scooters and bicycles for their deliveries. They don’t transport those pizza boxes in semi-trucks. Using models is much the same thing—you have to choose the right tool for each purpose.”
Following the release of ChatGPT in late 2022, market enthusiasm blinded at least some companies, according to Hongisto. Instead of thinking about what problem they were trying to solve, companies conducted many AI experiments that didn’t necessarily have a clear goal. ”The management of many companies and organizations decided that they suddenly needed to do a massive amount of things with AI. At that point, many forgot that even though AI is a uniquely wonderful tool, it is ultimately just a tool.”
Gyursanszky points out that many companies were already involved in developing their own systems—based on machine learning, for example—before ChatGPT was released. Since then, a lot of money and resources have been spent on things that did not turn out well. ”I believe we have certainly wasted a lot of energy and money on language models, but we have also certainly learned a lot.”
The US government has taken a harder line on curbing AI—especially after the release of Anthropic’s Mythos model, which effectively finds software vulnerabilities—but at the corporate level, Hongisto views American companies as more willing to see the benefits of AI in the big picture. ”In Finland especially, there have been economic challenges, so AI has been viewed as an excellent way to cut costs and improve productivity. And while in many cases it is, I think it’s thinking too small if you only focus on minor cost-pinching.”
Investments can still overflow if the necessary guardrails are not built into the products. For example, Uber reported that it spent its entire annual AI budget in just four months. Hongisto emphasizes that mistakes or irresponsible actions occur in these types of situations. ”There have been cases in the headlines where people rushed in with immense enthusiasm without any control over who gets to burn how many tokens, and so on. The mistake happened when they failed to realize that control must be maintained.”
In enterprise use both Hongisto and Gyursanszky state that they consider how AI is used to be more important than blindly staring at whether a company uses a language model from OpenAI or Anthropic. Additionally, training data comes to the forefront. ”Agents only work as well as the data available to them. That’s why we need connections to different data sources and consideration of what data each agent can use, because not everyone needs access to the entire company’s dataset,” Gyursanszky explains during the demo.
Legal issues can also arise with data sources, which is why it is good for management to be aware of what the AI can access and what it can do with that data. ”For example, the board and the CEO have massive legal responsibilities in companies. It is no laughing matter what AI agents get to do in their name,” Gyursanszky elaborates.
Both emphasize that development in most companies starts with individual people and teams. There can be years of difference in AI skills and understanding among employees. A key problem is that understanding should be strong, particularly within the executive team, but in reality, this is often not the case. ”Executive teams should be more informed and not outsource AI development solely to IT or the Chief AI Officer. When there is a strong understanding within the leadership team of how agents are used and built, the company can start planning a path that truly changes its direction,” Gyursanszky summarizes.
Hongisto adds right after that this doesn’t apply to all executives, noting there are cases where management can be a bit too enthusiastic. ”I once heard a CTO joke, asking if we could build a platform where our CEO could ‘vibe code’ in a more controlled way?” Hongisto says. According to the CTO, the CEO was becoming ”a bit too active.”
At Google, management has condensed development into two things: encouraging users and leading by example. Hongisto says that making a bit of a fool of himself has been a good way to get employees excited about using AI. He considers experimenting and failing in front of others to be his personal superpower. ”Sometimes it’s even better for management to do something and fail in front of everyone, because then the team sees that, ‘Well, at least I can do better than Markus,’” Hongisto laughs.
Company Profile: Codento
- What: An IT consulting company specialized in Google Cloud technologies. Codento operates in Finland, Sweden, and Norway.
- CEO: Anthony Gyursanszky
- Founded: 2005
- Revenue: €10 million (proforma)
- Employees: 75
- Other: Google Cloud awarded Codento with the Partner of the Year 2026 award in Finland.
Read the article in Kauppalehti (in Finnish): Kauppalehti