Glossary
Not Wikipedia definitions. A glossary grounded in practice, with an opinionated editorial perspective.
Architecture
MCP server A server implementing Anthropic's Model Context Protocol, which exposes tools and resources that an LLM can use in its workflows. Multi-agent Architecture where multiple specialized AI agents collaborate or relay on the same task or project. System prompt Persistent instructions given to an LLM before any user interaction — the frame within which the agent reasons. Tool use An LLM's ability to call external functions (tools) during generation, to execute actions or retrieve data.
Tooling
Coding agent A software tool that combines an LLM with access to the development environment to execute coding tasks autonomously. Vendor lock-in (AI) Dependency on an AI provider to the point where switching tools or models carries a real migration cost: restricted subscriptions, non-portable configs, frozen workflows.
Models
Context window The maximum amount of text (in tokens) an LLM can process at once, including conversation history, instructions, and loaded files. LLM Large Language Model — a language processing model trained on massive amounts of text, capable of generating coherent text and reasoning about instructions. Token The processing unit of an LLM — approximately 0.75 words in English. LLM API pricing is calculated in input and output tokens.
Workflow
Agentic workflow A working mode where an AI agent autonomously executes tasks over an extended duration, with access to files, tools, and the development environment. AI enablement The work of structuring and spreading AI practice across a team: versioned conventions, shared commands/skills, documented tool choices, onboarding. Context rot The progressive degradation of an LLM's response quality over a long session, caused by the dilution of initial instructions within the context. Model routing A policy of directing each task to a different AI model based on risk, complexity or cost, instead of using a single model for everything. Prompt engineering The art of formulating instructions so an LLM produces exactly the desired result — not approximately. Shadow AI Adoption of AI tools or accounts not validated by the organization — personal accounts, prompts containing proprietary code, unaudited extensions.
Cost
AI ROI Measuring the real return of agentic practice through honest proxies (session duration, commit ratio, tokens saved) instead of vanity metrics (number of prompts). Inference cost The dollar cost of running an LLM to generate a response — calculated in input and output tokens, varying by model and provider.