How do AI agents create value in Norwegian companies? That was the topic when AIKI gave a four-hour guest lecture at BI Norwegian Business School for students in Business Development and Digitalisation.
These students enter Norwegian companies shortly, so we spent the four hours on the question they will actually meet there: what does it take for an AI agent to do a real work task, every day, with a result someone can rely on?
The answer is three things. They are undramatic, and they decide whether an agent ends up in production or stays a demo.
1. Good data sources
An AI agent is only as good as what it is allowed to see. This is the part most often underestimated, because it is about integration and access rights. Model choice comes much later.
An agent that qualifies incoming leads needs access to the CRM, to the history of previous customers, and to which deals actually closed. Give it only the email that arrived, and it guesses. It guesses confidently.
In practice the job starts by mapping where the data lives, who owns it, and what the agent should be allowed to read and write. It is dull work, and it decides the outcome.
2. Precise instructions
The second mistake is writing the instruction as a wish rather than a rule.
"Assess whether this enquiry is interesting" gives a different answer every time. "Mark as qualified if the company has more than seven employees, is based in Norway, and the enquiry mentions a specific process" gives the same answer every time, and it can be checked when someone disagrees.
The instruction should state what the agent must do, what limits it has, and what it should do when uncertain. That last part is almost always forgotten. An agent with no definition of doubt will make a decision regardless.
3. Well-considered routines
The third part is the operation around the agent. When does it run? Who sees the result? What happens when it gets something wrong?
Irreversible actions go through a human. If the agent emails customers, moves money or deletes data, it drafts and a person clicks. Anything else is a risk waiting to happen.
And the agent is measured on the outcome: did the recipient get what they were supposed to get? A job marked complete is a signal that the code ran. The proof sits at the other end, with whoever was meant to receive the result.
From lecture hall to production
The method we taught at BI is the one we use when we build AI agents for Norwegian companies. It holds in both rooms, because the constraints are the same: the data is scattered, the instructions are imprecise, and someone has to decide what happens when the agent gets it wrong.
AIKI builds custom AI agents and complete AI solutions through AI Development, and trains employees to use them through AI training. The combination is the point: an agent the staff understand survives its first quarter.
Want to explore what AI could do for your business? Get in touch.
