liveBook · Manning

Cognitive module

As explained in Designing AI Agents

A separable agent subsystem responsible for a cognitive function such as perception, memory, reasoning, action, reflection, or collaboration.

A cognitive module is a separable agent subsystem responsible for a design concern, such as perception, memory, reasoning, or action. Use the modules as a checklist: a task may require some capabilities lightly and others deeply, depending on its constraints and risks. section 2.1.1

The capability stack has three layers. Core cognition comprises Perception, Memory, Reasoning, and Action. Reflection adds a feedback path that can change current work or future behavior. Collaboration and Governance address coordination, authority, and production boundaries. The stack is not a maturity ladder: every added responsibility needs controls and evidence, and carries operating cost. section 2.1.1 section 2.7

Table 2.1 The seven cognitive modules of an agent system
Module
Core Question
Chapter
Perception
What does the agent see, and what does it pay attention to?
3
Memory
What does the agent remember, and how does it retrieve knowledge?
4
Reasoning
How does the agent think, and how deeply?
5
Action
What does the agent do in the world?
6
Reflection
What evidence evaluates the work, and what may that feedback change?
7
Collaboration
How do agents (and humans) work together?
8
Governance
How is the agent constrained, observed, and kept aligned?
9
 
 
 

Use the seven core questions to identify which responsibilities the task requires. section 2.1.1

Modules become an architecture through their information flows. Perception filters and structures the external information available to downstream modules; poor filtering can affect every later step. Memory supplies shared state and evidence. Reasoning combines context and prior experience to choose a plan, and Action turns the proposal into an external interaction. A successful tool return is evidence of execution, not by itself proof that the intended effect was accepted. section 2.2.1

Reflection may propose lessons for Memory, but only admitted lessons may influence live or later work. The runtime hosts the modules and mediates access to the external world; governance controls can reach every module. Make state, tool calls, permissions, and evaluation results observable, and verify intermediate results to limit compounding errors. section 2.2.1 section 2.2 section 2.7

Controlled rollout

For a rollout change, Perception finds an existing utility, Memory retains the team's rule, and Reasoning chooses whether to reuse, extend, or replace it. Action defines permitted code and configuration changes. Reflection checks the result against the rule and feature request; Collaboration supplies an independent view. Governance determines who can increase exposure, what evidence permits it, and how to stop the rollout. section 2.1.1

How Designing AI Agents covers it

How deep it goes
Defines seven responsibilities and their information flows, while deferring detailed pattern alternatives.
Examples
Applies all seven modules to a controlled-rollout change involving code, configuration, independent review, and exposure controls.
What you can do after
Evaluate and understand which agent capabilities, boundaries, and failure controls a task requires.
How it connects
Connects cognitive modules to the Perception-Reasoning-Action loop, runtime infrastructure, feedback, and governance.

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.