Inside a LLM — Transformers, Attention & Reasoning
From tokens to generation, the anatomy of a language model — hands-on, with no equation before the intuition that earns it.
AI Blog
A Journey Inside AI Agents
Understand what happens between your prompt and an action in the real world.
Five chapters, each one layer deeper into the machine. The figures are manipulable: you move tokens, weigh attention, watch a decision pass a policy. No prior knowledge assumed.
The journey
Each layer of the diagram is a chapter.
The chapters
When does an LLM become an Agent?
Tools, state, loop, autonomy — what has to be built around a model before generation becomes action, and what that costs.
RAG — Giving an AI a Library without Re-training it
Chunk, represent, retrieve, augment — the pipeline that gives a model context, and the five ways it produces a wrong answer without warning.
MCP — Giving an AI Hands
Host, client, server, and the three things a server exposes. The protocol that standardises access — and the question it leaves wide open.
Why Agentic AI Needs a Control Plane
Identity, policy, human approval, trace, evidence — the layer missing between an agent's decision and its effect on the world.