Almanac

Learning path

The Agent and Tool Ecosystem

How do AI models go from answering questions to actually doing things? This path traces the agent and tool ecosystem — from the platforms and protocols that connect models to the world, to the coding agents and model families that put it all into practice. Designed for readers who want to understand not just individual models, but how the whole system fits together.

Mixed level9 steps~52 min

9 steps

Begin →

  1. Model Context Protocol

    Start here: the Model Context Protocol is the connective tissue of the agent ecosystem — understanding how models talk to tools makes every later step click.

    Read →BeginnerIn-depth

  2. Hugging Face

    Hugging Face is the open platform where models, datasets, and tools are shared — the marketplace the ecosystem runs on.

    Read →BeginnerIn-depth

  3. OpenAI

    OpenAI pioneered the API-first model that made tool-calling and agent frameworks possible, and its design choices still shape the field.

    Read →BeginnerIn-depth

  4. Codex

    Codex was the first widely-deployed model to turn natural language into executable code — the proof-of-concept that launched the coding-agent wave.

    Read →BeginnerIn-depth

  5. GPT-5.5

    GPT-5.5 represents where OpenAI's tool-using, agentic capabilities stand today — a concrete anchor for the abstract ideas above.

    Read →BeginnerIn-depth

  6. Anthropic

    Anthropic's safety-first approach to building agents offers a useful contrast to OpenAI's, and sets up the Claude family that follows.

    Read →BeginnerIn-depth

  7. Claude

    Claude is Anthropic's flagship model line — understanding its design philosophy shows how safety constraints shape what an agent can and will do.

    Read →BeginnerIn-depth

  8. Claude Code

    Claude Code is the agentic coding product built on top of Claude — the clearest example of how a model family becomes a real-world tool-using agent.

    Read →BeginnerIn-depth

  9. Mistral AI

    Mistral AI rounds out the picture as a leading open-weight alternative, showing how the agent ecosystem extends beyond the two dominant closed-model labs.

    Read →BeginnerIn-depth