Agent Anatomy
4 components: Brain (LLM) + Tools + Memory + Perception. Know how each works.
Brain (LLM): the decision-maker. Reasoning, planning, tool selection happen here. Model choice (Claude / GPT / Gemini / Llama) shapes the agent's character.
Tools: the bridge to the external world. Web search, code execution, file I/O, API calls, DB queries. Each tool = schema + execution function + result formatter.
Memory: two levels — (1) Short-term = conversation history, recent observations. (2) Long-term = vector DB, episodic memory, knowledge base. The longer the agent runs, the more critical memory becomes.
Perception: input modality — text, image (vision), audio, structured data, screenshots (computer use). Modern agents start multimodal.
Orchestrating these 4 components = agent loop.
What you'll gain
Looking at any agent, you can identify the 4 components and spot what's missing.