Cognition is the quality of an agent’s internal model of a task: how it represents what is being asked, decides how to proceed, and updates its understanding as evidence arrives. It describes observable engineering capabilities, not machine consciousness. Reasoning patterns help solve well-defined problems; cognition helps identify what an ambiguous problem requires, while metacognition monitors whether the agent has understood it. section 10.1.3 section 10.1.1
Four capabilities make up this task model:
- Task decomposition breaks a novel problem into subtasks without relying on a supplied template. section 10.1.3
- Dependency reasoning identifies which subtasks rely on others and which can run in parallel. section 10.1.3
- Compositional tool use combines tools in new configurations to fit the problem. section 10.1.3
- Model updating revises the agent’s understanding or plan when new evidence contradicts its initial hypothesis. section 10.1.3
From separate lookups to a composed task
An agent may be able to perform two lookups separately yet fail to join or compare their results. Cognition means recognizing that the request needs multiple steps and planning a tool sequence to complete them. section 10.1.1 section 10.1.3
A cognitive agent architecture organizes these abilities around a shared workspace, with perception, planning, execution, evaluation, and attention modules. A memory module can add persistent experience, retrieving past strategies before planning and recording outcomes afterward. section 10.6
Table 10.1 Agent cognitive capabilities
| Capability |
Present |
Absent |
Supporting primitive |
| Task decomposition |
Agent invents novel subtask sequences |
Agent treats complex queries as flat lookups |
CoT, planning |
| Dependency reasoning |
Agent sequences steps while respecting dependencies |
Agent executes steps in arbitrary order |
ToT, sequential thinking |
| Compositional tool use |
Agent chains tools in novel configurations |
Agent maps queries to single tool calls |
ReAct |
| Model updating |
Agent revises plan when evidence contradicts it |
Agent follows initial plan regardless |
Reflexion, replanning |
Compare each capability with its absent counterpart and the reasoning primitive associated with it. section 10.1.3
