Accuracy
Correct outputs
Metrics
Speed
Response time
Cost
Generation cost
Reliability
Consistency
Methods
Manual review
Human QA
Automated testing
Automated checks
Benchmarking
Compare against baselines
User feedback
Collect feedback
Frequently Asked Questions
What agents can be evaluated?
Any text/data output agents
How often to evaluate?
Regularly, especially after updates