constraint handling

strict hooks to make sure it doesn't order anything too expensive" — Demonstrates implementing safety constraints and guardrails to limit agent behavior and prevent undesired actions (excessive spendi

no tool names anywhere in the input" — Demonstrates how the model adapts its behavior under strict constraints (absence of tool name information), revealing constraint-based decision patterns

[article] "Current models lack deep causal or conceptual understanding" — Article identifies lack of causal and conceptual understanding as a core limitation of present-day AI models