Why we don't sell agents
Every vendor wants to sell you an agent. We learned the hard way that it is a bad product for an organization that isn't ready for it.
I watched dozens of agent projects in large organizations up close. Most looked great in the demo. Most did not survive the first quarter.
The reason was not the technology. The reason was that the agent entered an organization whose people did not know how to work with the tool underneath it. They didn't know what to ask, couldn't spot a mistake, couldn't fix one. When something broke, they went back to Excel.
The right order
First the people learn to work with the tool. Not in a lecture. In small working groups, on their own examples. The procurement manager on her orders. The shift manager on his logs.
After two weeks, three things happen on their own. Employees build small skills by themselves. They discover where the connection to the systems is missing. And they can say precisely what they want automated.
Only then do we harden. We take what already works by hand, and make it run alone, with human approval before every write to a core system.
Why this is better for us too
An agent project ends. The client pays, receives, and we leave. If it stalls, that's their problem. That is not a business we want.
An AI department doesn't end. We measure ourselves by how many employees use it, not how many bots were delivered. When employees build alone, we move to the next department.
Which tool did you buy, and how many people open it in the morning?