Fundamentals
What makes a system 'agentic' rather than a plain LLM call. The four-component reference architecture, memory types, the perception–reasoning–action loop, and the open problems no framework solves.
Agentic systems are LLM-powered programs that perceive, reason, and act in a loop — autonomously enough to handle multi-step work, but with their behaviour bounded by the tools and instructions you give them. The Fundamentals topic covers the small set of ideas that make every other topic make sense: what an agent actually is, how its architecture decomposes, where memory fits, how perception and action wire to the model in the middle.
Most confusion about 'agents' is upstream of any framework. People reach for LangChain or AutoGen before they've internalised what an agent does differently from a chat call. These pages exist to slow that step down: read this topic before evaluating tools.
Key concepts
- An agent is a loop, not a single call — perceive, reason, act, observe, repeat
- Four-component architecture: model + tools + memory + instructions. Drop any of them and the system breaks differently
- Memory is not one thing — short-term context, episodic, semantic, procedural all serve different purposes
- Grounding is the bridge from raw input to actionable representation — text, vision, audio all need it
- Most failures aren't model failures — they're loop-control, memory, or tool-design failures
Reference template
// The four-component agent
┌────────────┐
│ Model │ ◄── reasoning, planning, decisions
└─────┬──────┘
│
┌─────┼──────┐
│ Tools │ ◄── perception in / action out
└─────┬──────┘
│
┌─────┼──────┐
│ Memory │ ◄── short-term + long-term state
└─────┬──────┘
│
┌─────┼──────┐
│ Instructions │ ◄── role, rules, format, safety
└────────────┘Adapt to your problem; the structure is the load-bearing part.
Common pitfalls
- Treating an agent as a glorified chat — agents act in the world, chat models don't
- Skipping memory design — every long-running agent eventually hits a context wall
- Underestimating perception — bad grounding silently caps every downstream capability
- Conflating 'autonomous' with 'unsupervised' — autonomy without observability is just a black box
Related topics
Items (7)
- What Is an AI Agent?
What makes a system 'agentic' vs a plain LLM call. Autonomy, tools, memory, and the perception–reasoning–action loop.
Concept Foundational - Agent Architecture Overview
The four-component reference architecture: model, tools, memory, instructions. How requests flow through each.
Concept Foundational - Agent Memory
Short-term working context vs long-term storage. Episodic, semantic, and procedural memory in agent systems.
Concept Foundational - Perception and Grounding
How agents take input — text, vision, audio, structured data — and ground it to actions in their environment.
Concept Foundational - Reasoning and Planning
Chain-of-thought, tree-of-thought, plan-then-execute. Where the agent's 'thinking' happens before it acts.
Concept Intermediate - Action and Tool Use
How agents act in the world. Function calls, code execution, API requests, and the safety boundary around each.
Concept Foundational - Key Challenges in Agentic Systems
Hallucination, long-horizon drift, cost overruns, evaluation difficulty, prompt injection, and the open problems no framework solves for you.
Concept Intermediate