Blast radius is the scope of damage from one bad run. It's the right frame because it forces a concrete answer where 'what could go wrong' invites a vague one.

The Four Questions

1. WHAT IS THE WORST SINGLE ACTION IT COULD TAKE?

Not the worst you can imagine. The worst reachable

through the tools you've actually given it.

2. HOW MANY TIMES COULD THAT HAPPEN BEFORE ANYONE NOTICES?

Once, because a human approves each one? Or four

hundred, because it runs overnight on a queue?

3. HOW WOULD WE FIND OUT?

A customer complains? A monitor fires? Nobody,

until the monthly review?

4. HOW FAST CAN WE STOP IT?

Is there a switch? Who can press it? Do they know

they can? Is it documented anywhere findable at 3am?

Question 2 is the one that turns a small problem into

an incident. A single wrong email is an apology. Four

hundred wrong emails is a very different week.

Rate Limits Are a Safety Control

The cheapest blast-radius reduction available is a cap on how many irreversible actions an agent can take per hour. It costs almost nothing to implement, it rarely affects legitimate use, and it converts a potential mass incident into a small one. Ask for it explicitly, it's often not there by default.

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

The Blast Radius Question