10 Exploring the cognitive agent that thinks, monitors, and adapts

At this point, we’ve now built agents that can reason (CoT, ReAct, ToT), reflect on their mistakes (Reflexion), iterate toward goals (the agentic loop), and remember past interactions. Since each of those capabilities is powerful on its own, you might think that stacking reasoning patterns on top of each other will make an agent smarter. But the truth is, stacking them together just makes the agent more complicated. Think of the collection of agents we’ve built like a toolbox full of dull instruments, with no craftsman to decide which to pick up and use. This chapter is about building the craftsman.

10.1 Understanding agent cognition and metacognition as engineering concepts

10.1.1 The five failure modes of capable-but-not-cognitive agents

10.1.2 From reasoning primitives to cognitive architecture

10.1.3 Defining cognition for agents

10.1.4 Defining metacognition for agents

10.1.5 Three theoretical foundations

10.2 Mapping the mind into a cognitive agent architecture

10.2.1 Architecture overview

10.2.2 The Cognitive Workspace

10.2.3 The Perception Module

10.2.4 The Planning Module

10.2.5 The Execution Module

10.2.6 The Evaluation Module

10.2.7 The Attention Module

10.2.8 The Memory Module and the MCP memory server

10.3 Building and running the Cognitive Agent

10.3.1 The cognitive loop

10.3.2 A complete cognitive agent with MCP

10.3.3 Walkthrough: watching the cognitive cycle in action

10.3.4 Confidence-gated execution

10.3.5 Stagnation detection and strategy pivoting

10.3.6 Knowledge boundary awareness

10.3.7 Emergent behaviors

10.4 Measuring cognitive capability and looking ahead

10.4.1 Cognitive efficiency metrics

10.4.2 Before and after: measuring the impact

10.4.3 The road to more general agents

10.5 Exercises

10.6 Summary