liveBook · Manning

Metacognition

As explained in Designing AI Agents

Meta-cognition is the ability of an agent to evaluate and improve its own cognitive activities or outputs.

Meta-cognition adds Reflection to an agent’s core cognition: feedback examines an artifact or outcome and may change current work or future behavior. It makes review explicit after a first-pass output. section 2.1.1

Treat reflection as a capability, not a maturity badge: adding it brings controls, evidence, and operating costs that the task must justify. section 2.1.1

Coding-task results

Reflexion-augmented agents improved from 50% to 90% success on coding tasks, a reported result rather than a guaranteed outcome. section 2.1.1

How Designing AI Agents covers it

How deep it goes
Defines reflection as review of first-pass outputs and reports coding-task results, but does not specify an implementation procedure.
Examples
Uses coding tasks as a reported case, where Reflexion-augmented agents improved from 50% to 90% success.
What you can do after
Evaluate first-pass agent outputs through an explicit review step that may revise current work or future behavior.
How it connects
Connects Reflection to Perception, Memory, Reasoning, Action, Collaboration, and Governance in the agent capability stack.

Also explained in

Compare how 3 books explain metacognition →

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.