In short
Grading the work. Measuring whether an agent's output meets the bar, so changes can be tested instead of hoped for.
Evaluation is how you know an agent is doing its job. A set of real cases with known correct results, run through the system, scored. Did the post pass the layout rules? Was the lead classified the way a person would classify it? Did the brief include the number that matters?
It is what makes change safe. Swap the model, tighten a rule, add a step: run the evaluation, compare. Without it every change is a guess. With it, improvement is measurable and regressions are caught before a client sees them.
Validators are the automated, always-on half. Evaluation is the deliberate half, run when something changes. Both are written down.
Example
What it looks like.
Twenty past drafts with their approved published versions. After a model upgrade, all twenty run again. Eighteen match or improve. Two get looked at by a person.Related
Read next.
- validatorA script that refuses bad output before it ships. The agent does not get to decide whether its work is good enough. The validator does.Read the definition
- output is informationA bad result is data about what the model understood. Read it, adjust, go again. Beginners stop at two tries. Builders go fifteen.Read the definition
- modelThe intelligence an agent calls when it needs to read, write, or decide. Replaceable. When a better one ships, the same instructions run on it and the results improve.Read the definition