Why AI Agents Don't Move Beyond Pilots
The main reason I see AI agents not moving beyond pilots is not the agent itself.
It's how the pilot is framed.
The Technology Works Fine
In most cases, the technology is built to execute a specific task. The task works. The demo looks fine.
But companies are not isolated systems.
The Real Challenge
What I see at Prosus, and across large organizations more broadly, is that work lives in a mix of:
- Written processes
- Unwritten behavior
- How work actually moves through the organization
- Where decisions get checked
- Where trust is required before an output can be used
When an agent is designed in isolation, it fits the task, but not the organization around it.
That mismatch rarely shows up in a pilot. It shows up when teams start trying to rely on it.
The Hidden Constraint
There's another constraint that often goes unnoticed.
Many operational flows are still shaped by technology decisions made 10, 15, sometimes even 20 years ago. Pilots inherit those boundaries, even when the agent itself is new.
At that point, the issue is no longer:
- Model quality
- Tooling capabilities
It becomes a question of organizational knowledge:
- How that knowledge is embedded in systems, roles, and decision rights
- Whether the agent is treated as something strategic, or just something that exists
From Experiment to Strategic Capability
That's also the point where an agent stops being an experiment and starts becoming a strategic capability.
For me, that's usually the point where the real work starts.
What have you seen block AI agents from moving beyond pilots? I'd love to hear your experiences.
Originally published on LinkedIn