Welcome Part 2 of Summary of Agentic Design Patterns by Antonio Gulli.
Planning & Goal-Driven Behavior
True intelligence requires foresight. The Planning pattern enables agents to formulate action sequences from an initial state to a goal state—not following a script, but dynamically charting a course. When obstacles arise (venue unavailable, API down), capable agents don’t fail; they adapt, re-evaluate constraints, and find new paths. Combined with Goal Setting and Monitoring, agents track their own progress against success criteria, knowing when they’ve achieved objectives and when to pivot. This is AI that doesn’t just execute—it strategizes.
Memory Management – Short-Term vs Long-Term
Humans remember. Agents need to as well. Short-term memory lives in the context window—recent messages, tool results, and current task state. But context windows are limited, so agents must strategically summarize and prioritize what stays. Long-term memory persists across sessions using vector databases and retrieval mechanisms, allowing agents to recall past conversations, learned preferences, and accumulated knowledge. The magic happens when both work together: an agent that remembers your communication style from last month while staying focused on today’s task.
Multi-Agent Collaboration & Communication
Complex problems demand specialized teams. The Multi-Agent pattern structures systems as cooperative ensembles—a Research Agent retrieves information, a Data Analyst processes numbers, and a Synthesis Agent compiles the report. But specialists need to talk. Enter Inter-Agent Communication (A2A)—Google’s open protocol enabling agents built with different frameworks (LangGraph, CrewAI, ADK) to coordinate, delegate, and share context. The result: systems where the whole dramatically exceeds the sum of its parts.
