The real bottleneck in scaling operations isn’t automation. It’s capture.

Enterprise software is moving from systems of record to systems of action. The premise is that AI agents can read a business record, make a decision, and act without waiting for a person.

But there is a more fundamental question: Was the operation accurately captured in the first place?

For desk-based work, that may be a reasonable assumption. In warehouses, plants, fleets and field operations, it often is not. Many platforms expect someone to stop working, find the record, and describe what happened. Under operational pressure, that moment may never arrive.

Over the past quarter, I conducted discovery interviews with operators at 30 organizations across eight industries. The environments differed, but the same limitations repeatedly surfaced. Equipment was taken before it could be checked out. Devices due for calibration could not be confirmed as returned. Work was completed physically but represented inaccurately in the system.

This is not simply stale data. It is a capture gap: parts of the operation never enter the system meant to represent them.

Handing that incomplete record to an agent can make the problem worse. The agent cannot distinguish a verified fact from a number entered simply to close a task unless the system provides evidence of the difference.

The fix has to begin before automation. Software must capture work as it happens: through a scan while a tool is being collected, a photo documenting damage, a voice command when both hands are occupied, or a GPS signal that requires no manual entry.

The principle is straightforward: capture first, verify second, and automate once both are in place.

Before asking what an agent can do, ask whether the operation it will oversee was accurately captured. No agent can account for a gap that was never shown.

Salman Abubakar, EZO principal product manager