Role Mismatch Is More Common Than Lack of Ability
When an agent underperforms, the first reaction is almost always the same sentence: the model isn't good enough — swap in a stronger one.
Occasionally that's right. Usually it isn't. The more common cause is not lack of ability but role mismatch — a JD that was never written clearly, responsibilities that fight each other, a role used in ways it wasn't defined for. Swapping models mostly treats the symptom; redefining the role treats the disease.
The Standard Faces of Mismatch
Role mismatch wears a few fixed faces.
One is a role that's too wide. A single agent plans, executes, and reviews — three responsibilities that belong apart, crammed into one JD. So it second-guesses itself as it works: planning gets interrupted by execution, and review goes soft on its own output. That is not a model lacking intelligence. That is one person handed three jobs that fight each other.
Another is usage that contradicts the role. The JD defines a narrow position, but at call time the agent is treated as a generalist and handed everything. Or the reverse: the work needs it to own a domain continuously, yet it's invoked as a disposable tool every time, with no memory and no environment.
A third is missing boundaries — both overreach and inaction start there.
What these failures share: the symptom shows up in the agent's behavior, but the disease lives in the role definition. Tuning the model while staring at the behavior is treating the symptom.
Fix the JD First, Then the Model
So there's an order of operations: when debugging an agent, fix the JD first, then the model.
When behavior is wrong, go back to the role — are the responsibilities too broad, are the boundaries undrawn, does the usage contradict the definition? Straighten those out and many "the model isn't good enough" problems disappear on their own. Only when the role is genuinely in order and behavior still falls short is it time to suspect the degree and go find a stronger candidate.
A degree is not job competence; mismatch is more common than lack of ability. Both point to the same word: fit. First a clear role, then the right hire. The main battlefield of agent engineering is role definition, not model selection.