Decode

An AI system that can take actions autonomously — not just answer questions, but actually do things: browse the web, write code, call APIs, manage files.

Adjective describing AI behaviour that is goal-directed and multi-step — the AI decides what to do next rather than just responding to a single prompt.

Command Line Interface — you type text commands directly into a terminal instead of clicking around a GUI. How developers talk to computers (and increasingly, to AI tools).

How much text an AI can 'hold in its head' at once — its working memory. Bigger = more expensive but smarter in long conversations or large document tasks.

Training an existing model further on your own data to make it better at a specific task or style. Expensive, often overkill — RAG is usually the right answer first.

When an AI confidently states something false — it doesn't 'know' it's wrong. The #1 reliability problem in production AI products.

Integrated Development Environment — a fancy text editor for writing code. Think Xcode but for everything. Cursor and VS Code are the popular ones right now.

Model Context Protocol — Anthropic's open standard for connecting AI models to external tools, data sources, and services. Like USB-C but for AI integrations.

The actual AI brain — the trained neural network that processes input and generates output. GPT-4, Claude 3.5 Sonnet, Gemini are all 'models'.

The craft of writing instructions to AI models to get reliably good outputs. More art than science — but there are real patterns that work.

Retrieval-Augmented Generation — instead of asking the model to 'remember' facts, you fetch relevant docs at runtime and inject them into the prompt. Smarter than fine-tuning for most use cases.

The unit of text that AI models process — roughly ¾ of a word. API costs are measured in tokens. 1,000 tokens ≈ 750 words. This is the 'spend' on those Instagram dashboards.