Tech breakfast: Now it's about what AI agents can do for business

At the AI ​​agent meeting at Digital Norway yesterday, we got a good picture of where the agent field stands now, from conceptual understanding, via the public sector and governance, to startup, scaling and concrete construction.

Many still talk about AI as better text, faster searches and smarter assistants. That is not where the value lies. The value arises when AI connects to systems, data and processes – and performs work.

Alexander Haneng from Digital Norway set the scene well. A chatbot answers. An assistant helps. An agent acts. It can trigger processes, retrieve data, send email and work in the background. At the same time, the risk increases. Access to systems and data requires control, clear instructions and limited rights.

Kjetil Ringstad from the Norwegian Road Administration showed how this is implemented in a large company. Not as technology alone, but as structure and management. A model that worked well was how employees build their own agents, experts build joint solutions and developers build platforms.

The case with document control was concrete. Agents identify sensitive information, suggest changes and reduce manual work. At the same time, two principles are crucial: the handling of sensitive data must be controlled, and human in the loop must be in place before anything is published. This is where many encounter the real complexity.

Einar Michaelsen from Circular showed how this looks when it is connected directly to the value chain. Products can be digitized, categorized, priced and published in seconds. Sales agents find leads, score them and set the next step in the process. When 4 people work together with 40 agents, that says something about the direction!

The stack behind was, among other things, Linear, Claude Code, Cursor, GitHub, Supabase, Vercel, Gemini, etc. The value lies in how this is connected to a production line. For me, it was fun to see this, as I have experience with several of the tools and am now working on building my own agent solutions.

What distinguishes those who are having an effect now is how they start. Not with technology, but with concrete tasks. Repetitive work across systems, with clear rules and measurable value. A simple agent with limited access including human control is then built. When this works, it scales.

My take is simple. The question is not which models to use. The question is which work can be delegated, which processes should be redesigned and where this actually adds value.

It is also the background to the fact that I am now concretely working on building agents, with the ambition of a measurable effect in a business I collaborate with.

Inside the venue was a DeLorean with a "REBEL" license plate. In the film Back to the Future, it was about traveling forward in time. With AI, it feels more like we build it. AI agents are the way forward!