A cognitive agent builds a task representation, decomposes novel problems, reasons about dependencies, composes tools, and updates its plan when evidence changes. Cognition here means observable, engineerable capabilities rather than machine consciousness. section 10.1.3
A standard ReAct agent can execute a reasoning-and-action pattern, while a cognitive agent extends that pattern with novel task decomposition, dependency-aware sequencing, tool composition, and model updating. section 10.1.3 section 10.4.2
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 |
Use the capability table to connect each cognitive behavior with its failure mode and supporting primitive. section 10.1.3
Recognize cognition through four behaviors:
- Decompose unfamiliar problems into subtasks instead of treating complex queries as flat lookups. section 10.1.3
- Sequence dependent steps correctly and identify work that can proceed in parallel. section 10.1.3
- Chain available tools in configurations that were not explicitly demonstrated. section 10.1.3
- Revise a hypothesis or plan when new evidence contradicts the initial approach. section 10.1.3
Attention returns the agent to planning
In a cognitive cycle, perception classifies a query as complex, planning proposes HYPOTHESIS_TEST, execution returns low-confidence results, and evaluation signals REPLAN. Attention routes the cycle back to planning rather than advancing blindly. section 10.5
Memory improves later work
After one troubleshooting session records the problem, strategy, and outcome in a knowledge graph, proactive retrieval can surface that experience during a related second query and inform its plan. section 10.5 section 10.4.2
