A dedicated container on my Proxmox host runs a coding agent that writes, reviews, and ships changes to this site — through the same git-push pipeline I use. The interesting engineering isn't the model; it's the guardrails around it.
A language model generates one token at a time, and each token re-reads gigabytes of weights and a growing cache. Understanding the KV cache and the memory-bandwidth wall explains latency, context limits, and every serving optimization from the last two years.
Strip the frameworks away and an LLM agent is about 35 lines: call the model, if it asks for a tool run the tool and feed the result back, repeat until it stops. I built one from scratch to prove it — and the systems concepts underneath are all familiar.
Before the Model Context Protocol, every agent-to-tool integration was bespoke M×N glue. MCP is the USB-C of agent tooling: a standard client-server protocol so any model can discover and call any capability. What it is, and why it's a systems standard, not an AI one.
Before a language model sees your text, a tokenizer chops it into subword pieces from a fixed vocabulary — and that boundary explains the model's weirdest failures: bad arithmetic, the strawberry problem, why code indentation matters, and why some languages cost 3x more.