LinkedIn·Thursday, 6 August 2026·20d ago

I used to be very sceptical about AI agents, but this has changed a lot. These days, when working with Claude, it properly calls other…

Ben Van Delm

CCO at Horizon | Supply chain planning enthusiast | Decision Execution for Supply Chain Planning

I used to be very sceptical about AI agents, but this has changed a lot. These days, when working with Claude, it properly calls other tools, searches for information, can do research online, and can stick well to my prompt. It's not just a question-answer that is shielded from outside information, like it used to be. That loop: reason, act, observe, reason again, is what "agentic" actually refers to. It's a property of the system built around the model, not a change in what the model fundamentally is. Which is actually why the comparison to a travel agent holds up well, for instance. A good travel agent checks multiple sources, weighs what you told them you want, and takes action across several steps on your behalf. That's the loop. Whether each step is "just" next-word prediction underneath is irrelevant to whether the loop works. The same loop is now what's starting to run inside planning software: watching what a planner does, proposing the next action, checking the result. I don't think this is the end-all goal. In terms of optimization, for example, you don't want the system to learn from decisions that are suboptimal. You still want pure optimization. But it is definitely useful for other cases. #AI #SupplyChain #Agents

View on LinkedIn