Why Prompt Engineering Is a Smell

Why Prompt Engineering Is a Smell

When someone tells you they have “solved” AI with better prompts, pay attention to what they are really saying.

They are saying the system is unstable.

Prompt engineering exists because behavior is not locked.
It exists because outcomes depend on phrasing, context length, temperature, and upstream model changes.
It exists because the system cannot be trusted to behave the same way twice.

This is not craftsmanship.
It is compensation.

In mature systems, behavior is not negotiated at runtime.
It is designed, tested, and fixed.

You do not “prompt” a database to behave correctly.
You do not phrase requests carefully so a payroll system doesn’t misfire.

If AI requires constant linguistic tuning to remain useful, it is not yet infrastructure.
It is a prototype being mistaken for a system.

Prompts have a place—at the edges.
They do not belong at the core.

The more important a function is, the less it should depend on clever wording.

Browse topics Essay archive