Nederlands
Deutsch
At Commsverse in London - the community conference around Microsoft Teams, this year at Mercedes-Benz World - the first sessions kept circling one question: now that we are really using AI, how do we keep it affordable and under control? Two takeaways stuck with me.
AI adoption: from trying it out to managing costs
Organisations that have rolled out Copilot are now shifting from experimenting to managing usage and model choices. And that turns out to be hard, because the costs are difficult to predict.
One of the speakers compared it to giving a child 5 euros in a sweet shop - but never telling them when it runs out. And they come back with a 500-euro bill.
There is a second question hidden in there: is it really sustainable to let employees choose which model they use? For some that is freedom; for others it is mainly an extra burden - and for the organisation an open end on the invoice. Budgeting and governance are becoming the next big challenge around AI.
Agents: the next step, but still in their infancy
The direction was clear to everyone: agents are the next step. At the same time, most organisations have barely started. There is a gap between where things are heading and where companies stand today.
The best tip from the session was surprisingly concrete: treat an agent like a real employee. Give it a name, a role and a policy - and make sure you can let it go again if it does not do its job well. That keeps an agent from being an intangible piece of tech, and turns it into something with a role, responsibilities and boundaries.
My thread for the day: AI is past the experimentation phase. The value now lies in the discipline around it - control over costs, clear choices and governance that scales. Copilot and agents only really pay off when you set them up as seriously as the rest of your organisation.
