With a single-step AI feature, the output is the whole thing, if it's right, it worked. With an agent, the output is the end of a path, and a correct answer reached badly is a problem waiting to happen.

Outcome	Path	What It Means
Right answer	Took 4 sensible steps	Working as intended
Right answer	Took 19 steps and three wrong turns	Works but costs 5x and will time out eventually
Right answer	Guessed without checking anything	Got lucky. Will be wrong next time.
Right answer	Called a write tool it shouldn't have	A near-miss incident that scored as a pass
Wrong answer	Took the correct path, tool returned bad data	Not an agent problem, fix the data

Rows Three and Four Are Why This Chapter Exists

An agent that guesses correctly scores identically to one that investigates properly. An agent that took an unauthorised action but still produced the right answer scores as a pass. If you only evaluate outputs, both of these are invisible to you until one of them causes a real incident.

Agentic AI and MCP for Product Managers

Module 1What 'Agentic' Actually Means

Module 2How an Agent Actually Works

Module 3MCP: What It Is and Why You Should Care

Module 4Designing the Tool Layer

Module 5Permissions and the Human in the Loop

Module 6Evaluating Agents

Module 7Cost, Latency and Unpredictability

Module 8Designing What the User Sees

Module 9Risk and Blast Radius

Module 10When an Agent Is the Wrong Answer

Module 11Final Project: Design One Real Agent End to End

Module 12Appendix: The Prompt Library

Why the Final Answer Isn't Enough