28/10/2026
One small box on a warehouse wall was locating 823 assets on its own. If it failed, an entire zone would go dark.
Nobody had noticed. An AI agent caught it in its first week on the job.
Most asset tracking still works the same way. The system shows a problem, and someone has to notice it, dig through the data and decide what to do. With 30,000 assets and 2,800 anchors, nobody has time for that digging. So weak spots stay hidden until assets start going missing.
The agent didn't just flag the issue. It extended the anchor's scan time as a temporary fix, proposed two new anchors at that spot, and had the installation task ready for approval.
Here's what building agents for real operations has taught us:
- One agent, one job. A narrow task watched around the clock beats a general assistant waiting for a prompt.
- An insight without action is just another alert. One automotive supplier's agent pinpoints where engines get damaged in transit, matches it against ERP transport documents to name the likely carrier, and files the follow-up.
- Most agents don't need an LLM. Deterministic logic gives the same answer every time. Spend AI tokens only where they clearly add value.
Your tracking data probably already shows the next weak spot. The real question is how long it waits before someone looks.
Read the final part of a 4-part series by our CTO Johan Criel on AI and industrial asset tracking: https://eu1.hubs.ly/H0zns0t0
