Glossary
The terms, in plain language.
No jargon for its own sake. Just the words you’ll keep hearing, defined the way a practitioner would explain them. Filter, or link straight to a term.
- Agent
Software that decides its own next step in a loop, using tools and memory.
An AI agent is a system where a model runs in a loop: it looks at the situation, decides what to do next, takes an action with a tool, observes the result, and repeats until the goal is met. The defining trait is that the model chooses the steps. Contrast with a workflow, where the steps are fixed in advance.
- Workflow
A fixed sequence of steps. Predictable, but it cannot adapt.
A workflow runs a predetermined chain of steps. It may call a model at points, but the control flow is yours, not the model’s. Workflows are easier to reason about and test. Reach for an agent only when the path genuinely cannot be known ahead of time.
see:Agent
- Model
The reasoning engine at the center of the loop.
The LLM that reads context and produces the next action or response. In agent design, the model is one primitive among several — its quality matters, but so does what you feed it (prompting, memory) and what it can do (tools).
- Tools
Functions the model can call to act: search, code, APIs, files.
Tools turn a text generator into something that acts on the world. Each tool has a name, a description, and a typed input the model fills in. Well-designed tools (clear names, narrow scope, good errors) are one of the highest-leverage things you control.
- Memory
What the agent carries across steps and sessions.
Memory spans short-term context (the running conversation), summaries that compress old turns, and retrieved long-term facts. Managing it well is how agents stay coherent past a single context window.
see:Agent
- Prompting
The system prompt is code: role, rules, and tool usage.
The system prompt defines who the agent is, what it must and must not do, and how to use its tools. On serious projects it is versioned, reviewed, and tested like any other part of the codebase.
see:Model
- MCP (Model Context Protocol)
An open standard for connecting agents to tools and data.
MCP is a protocol that lets agents discover and call tools, read resources, and use prompts from external servers in a uniform way. Instead of hand-wiring every integration, you point your agent at MCP servers. It has become the common plug between agents and the rest of your stack.
- AGENTS.md
A file that gives coding agents durable project context.
A markdown file in your repo that tells a coding agent how the project works: conventions, commands, constraints. It is the persistent brief so you do not re-explain the project every session.
- Agent skills
Composable, reusable capabilities you can install into an agent.
Skills package a capability — instructions plus any helper files — so an agent can load it on demand. They keep the base prompt lean and let you add abilities (a writing style, a domain workflow) without bloating context.
see:AGENTS.md
- llms.txt
A machine-readable index that makes a site legible to AI tools.
A convention for publishing a clean, structured map of your site’s content for LLMs and agents to consume — the AI-native counterpart to a sitemap. This site ships one.
see:MCP
- Context engineering
Deciding what information the model sees, and when.
The discipline of curating the context window: which instructions, history, retrieved facts, and tool results to include for each step. On agent projects it is an architectural concern, not an afterthought.
- Agentic engineering
Directing a coding agent while owning the architectural decisions.
Building software (often agents themselves) with a coding agent, where you make the load-bearing decisions — boundaries, state, what the agent may not do — and review the choices embedded in the output. It is the disciplined sibling of vibe coding.