Skip to content

AI Hub β€” Agents and Agentic Systems

Agentic AI systems go beyond generating a single response β€” they plan, use tools, run loops, delegate to subagents, and take actions in the world. This hub groups the pages that explain how those systems work, how to design them, and what can go wrong.

  1. Spectrum of Agency in AI Systems β€” start here if you are new to agentic design. Maps the continuum from a simple prompt-response pair to a fully autonomous multi-agent system, with criteria for choosing where on the spectrum to operate.

  2. AI Agents β€” the building blocks: tools, memory types (in-context, external, episodic), control loops, and the basic ReAct and plan-execute patterns. Read before the agentic AI page.

  3. Agentic AI β€” higher-level orchestration patterns: how agents plan, how they hand off to subagents, human-in-the-loop checkpoints, and when agents fail. Assumes the foundations from the Agents page.

  4. Multi-Agent Pipeline β€” a worked production example: a release-notes pipeline with a research agent, a writer, a code checker, a scoring loop, and a human gate. Shows the full lifecycle from task ingestion to published output.

  5. RAG vs Agentic RAG β€” contrasts single-pass retrieval (simple RAG) with an agentic retrieval loop that plans, retrieves, checks relevance, and retries. Covers when the extra complexity is justified.

  6. OpenAI Codex (historical) β€” the original 2021 Codex model that established code generation as a mainstream capability. Provides context for understanding how modern coding agents evolved from it.

  • LangChain β€” Python framework for composing LLM calls, retrieval steps, and tool use into chains.
  • LangGraph β€” LangChain’s graph-based agent orchestration layer, designed for cycles, state machines, and human-in-the-loop workflows.
  • CrewAI β€” role-based multi-agent framework focused on task delegation and collaboration between specialized agents.
  • AI Integrations β€” patterns for integrating AI agents into existing applications.

After these pages, the Coding Agents overview shows how agentic patterns apply in practice across Claude Code, GitHub Copilot Agent Mode, and OpenAI Codex CLI. The Agentic AI Risks page covers what goes wrong in autonomous systems and how to design for it.