Tools fail constantly in production, records not found, permissions denied, services down, invalid input. What the tool returns in those moments determines whether the agent recovers sensibly or spirals.

Tool Returns	What the Agent Does	Verdict
Nothing, or an empty result	Assumes there's no data; may report 'no issues found'	Dangerous, silent failure
A raw technical error	Often retries the same call, or gives up entirely	Poor
'not_found: no shipment with reference 44182'	Checks the reference, or asks the user	Good
'permission_denied: needs ops_manager role'	Reports it and hands to a human	Good
'service_unavailable: retry in 30s'	Waits and retries once, sensibly	Good

The first row is the one that causes incidents. An agent investigating a delay whose lookup silently returns nothing may conclude there is no delay and tell the customer everything is fine. A tool that fails loudly and specifically is far safer than one that fails quietly.

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

What a Tool Returns When It Fails