liveBook · Manning

Accountability

As explained in Designing AI Agents

The ability to connect an outcome and the decisions that produced it to responsible agents and people.

Accountability connects an agent’s outcome and the decisions that produced it to the task, agent identity, and people responsible for the governing policy. An API log identifying the calling service is not enough, and natural-language chain-of-thought cannot substitute for execution evidence. section 9.1

For an agent action, retain evidence of:

  • The request the agent saw and the action and arguments it proposed. section 9.1
  • The policy decision that admitted the call, the tool’s result, and the resulting production change. section 9.1

Customer-service coupon

An agent issues a 100-yuan coupon. Accountability requires more than a record that its service account called the coupon API: a manager needs the proposal, policy decision, tool result, and production change. Authorization separately determines whether the proposal may proceed; containment limits the harm if an authorized decision is still wrong. section 9.1

How Designing AI Agents covers it

How deep it goes
Identifies the execution records needed for accountability but does not specify a logging implementation.
Examples
Uses a customer-service agent issuing a 100-yuan coupon to illustrate the evidence a manager would need.
What you can do after
Evaluate whether execution records connect an agent action to its task, policy, responsible people, and production outcome.
How it connects
Distinguishes accountability from authorization and containment in agent governance.

Also explained in

Compare how 17 books explain accountability →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Designing AI Agents. The section links point to the parts of the book it draws on.