100% Free Forever
AI-Powered Learning
Industry Expert Content
Certificates & Badges
Learn At Your Own Pace
Agent Evaluation & Observability
30 minadvanced

Why Agents Are Hard to Evaluate

Evaluating a chatbot that answers one question at a time is close to evaluating a classifier: you have an input, a reference output, and a scoring function, and you run it over a held-out set. Evaluating an agent breaks every part of that setup. An agent takes a goal, not a single question, and reaches it through a sequence of decisions it makes on the fly: which tool to call, what arguments to pass, whether to retry, when to stop. Two runs given the identical instruction can take different numbers of steps, call different tools in a different order, and still both be correct — or both be wrong in ways a simple answer-matching check will never catch.

This matters because the tools most engineers reach for first — exact string match, a fixed reference answer, a single pass/fail assertion — were built for deterministic programs and single-turn model calls. Applied to an agent, they produce two failure modes at once: they reject correct runs that took a different but equally valid path, and they pass incorrect runs that stumbled into the right final string by luck. Neither failure is rare. Both compound silently until a team ships a regression it never saw coming, because the eval it trusted was answering a question — 'does the output match?' — that was never the right question for a multi-step, side-effect-producing system.

The rest of this course builds a real evaluation and observability practice for agents: offline eval sets, trajectory-aware scoring, LLM-as-judge harnesses calibrated against humans, and the OpenTelemetry-based tracing that makes production agents debuggable instead of opaque. This first lesson does none of that engineering yet. Its only job is to make the underlying problem concrete enough that the rest of the course reads as a solution rather than as ceremony: agents are non-deterministic, multi-step, side-effecting systems, and cost and latency are outcomes, not footnotes.

Analogy🏏Cricket
🏏 Think of it like cricket: Chasing 275 in an ODI final is not a single scripted path — it's an open problem with many correct solutions. In the 2011 World Cup final at the Wankhede, MS Dhoni promoted himself up the order and finished the chase with an unbeaten 91, closing it out with a six over long-on. A different team facing the same target on a different day might get there with no boundary at all — just ones, twos, and careful running between the wickets until the required run rate flattens to zero. Neither approach is 'the' correct chase. What makes an innings a win isn't matching some reference scorecard ball for ball; it's reaching the target score, staying within the overs allotted, and not losing so many wickets that the chase collapses. A commentator who insists an innings is 'wrong' because Dhoni batted at four instead of at his usual number five, or because the winning stroke was a single instead of a six, is grading the wrong thing. The target score is the agent's task goal. The batting approach — which balls to attack, when to rotate strike, when to take the risk — is the agent's trajectory. Wickets lost are the side effects a costly wrong shot leaves behind. Overs remaining are the cost and latency budget. The final total is the only ground truth that matters, and there is more than one legitimate way to reach it.
Lesson 1 of 35
0% complete