Planner-Executor

Methods for breaking goals into steps; can be classical (A*, STRIPS) or LLM-driven with tool calls.

Extending agents with long-term memory stores.

Multiple agents interacting cooperatively or competitively.

Coordination arising without explicit programming.

Models evaluating and improving their own outputs.

Learning only from current policy’s data.

Balancing learning new behaviors vs exploiting known rewards.

Continuous cycle of observation, reasoning, action, and feedback.

Models trained to decide when to call tools.