What SAP support taught me about agent products
Enterprise software work shaped how I define problems before I reach for models.
Opening
Five years in SAP support trained a habit: separate the reported symptom from the actual job.
Main Argument
Agent products fail when they optimize for impressive generation instead of the user's next useful action.
Evidence or Example
In enterprise tickets, users rarely asked for "more intelligence." They asked for a reliable path through a broken process. The same pattern appears in football analysis and creative tools.
What Changed My View
I stopped treating model choice as the first decision. Problem framing and verification come earlier.
Practical Takeaway
Build the smallest loop that produces a trustworthy next step. Then decide which model can affordably support that loop.