Classical agents

Autonomous behavior can be built without machine learning. By combining state, rules, tools, and control loops, we can create programs that sense, decide, and act predictably. These patterns form the foundation of structured agent design.

• Long-running programs and state — Maintaining execution and managing program state.
• Deterministic decision logic — Rule-based action selection from explicit conditions.
• Tools as capabilities — Representing actions as callable functions.
• Workflows and control — Sequencing steps and handling branching logic.
• Agent loop — Implementing the sense–decide–act cycle.