Glossary
AI engineering, in plain English
The AI world runs on jargon, and most of it is simpler than it sounds. Here are the terms I actually use with clients, defined the way I would explain them on a call, so we can skip the buzzwords and talk about the work.
- Agentic Coding
- Agentic coding is when an AI does not just autocomplete a line but runs a whole loop on its own: reads your files, plans a change, edits several of them, runs the tests, reads the errors, and fixes what broke. The "agent" is the coding tool doing that loop with limited supervision. It is a real shift from an assistant that suggests to a colleague that acts, which is why guardrails and review matter so much.
- AI Engineer
- An AI Engineer is a software engineer who builds real products on top of large language models, not just someone who chats with one. The job is the usual engineering work of data flow, APIs, testing, and deployment, plus the new parts: prompts, context, retrieval, evals, and cost control. I use the label as shorthand for AI, Mobile, and Full-Stack Engineer, because in practice the AI layer sits on top of a normal, well-built app.
- AI Slop
- AI slop is the low-quality output you get when a model is left unchecked: generic filler text, code that looks plausible but is subtly wrong, duplicated logic, and the tell-tale phrasing that screams "a machine wrote this". "Anti-slop" is the discipline of preventing it, through clear rules, tight context, human review, and conventions that keep the model on rails. Most of my Claude Code setup work is anti-slop work, because the difference between a useful AI workflow and a mess is entirely in the guardrails.
- Claude Code
- Claude Code is Anthropic's command-line coding agent, the tool I use daily to build software with Claude directly in a real repository. It can read and edit files, run commands, execute tests, and work through a task across many steps rather than answering one question at a time. Setting it up well for a team, with the right rules and memory, is one of the productized services I offer.
- CLAUDE.md
- A CLAUDE.md file is the instruction sheet a coding agent reads before it touches your codebase. It captures your conventions, architecture, do-not-touch zones, and house style, so the AI follows your rules instead of inventing its own. A good CLAUDE.md is the single highest-leverage thing most teams can add: it turns a generic assistant into one that writes code the way your team already writes it.
- CLI
- CLI, or command-line interface, is how you drive a tool by typing commands in a terminal instead of clicking around a graphical app. It looks old-school, but it is where serious engineering and modern AI coding agents like Claude Code live: fast, scriptable, and easy to automate. When I set up a team's tooling, a solid CLI workflow is often what turns a slow, manual process into one an agent can run end to end.
- Context Engineering
- Context engineering is the practice of deciding what information an AI sees for a given task: which files, which docs, which examples, and in what order. Prompt engineering is about wording a single instruction well, but context engineering is the bigger, more durable skill of assembling the right surrounding material so the model can actually do the job. In real products the win almost always comes from better context, not a cleverer prompt.
- Context Window
- The context window is the amount of text a model can hold in mind at once, measured in tokens, covering both what you send in and what it writes back. Everything the model "knows" for a task has to fit inside it, so once you exceed the limit, older material falls out and the model forgets it. Managing this budget, deciding what to include and what to leave out, is a big part of building AI systems that stay accurate on large codebases and long documents.
- Fractional Engineer
- A fractional engineer is a senior engineer you retain part-time instead of hiring full-time, on a fixed monthly retainer. You get experienced hands and direction for the fraction of the week you actually need them, without the cost and commitment of a full-time salary. It suits founders and small teams who need real senior engineering, in AI, mobile, or full-stack, but do not yet have the volume of work to justify a permanent hire.
- LLM
- An LLM, or Large Language Model, is the kind of AI behind tools like Claude: a model trained on enormous amounts of text to predict and generate language. In practice it reads what you give it and produces a continuation, which is powerful enough to write code, summarize documents, answer questions, and drive agents. It is a probabilistic engine, not a database, so it can be brilliant and confidently wrong in the same breath, which is exactly why engineering around it matters.
- MCP
- MCP, the Model Context Protocol, is an open standard for connecting an AI model to outside tools and data sources like your files, a database, an issue tracker, or an API. Instead of every tool inventing its own glue, MCP gives them a common way to expose actions and information the model can use. It is what lets a coding agent safely reach beyond the chat window and actually do things in your systems.
- RAG
- RAG, or Retrieval-Augmented Generation, is a pattern where the system first fetches the relevant documents for your question, then hands them to the model to answer from. It is how you make an AI answer accurately about your own content, product docs, past tickets, an internal wiki, without retraining anything. Done well it cuts hallucinations sharply, because the model is grounded in real sources instead of guessing from memory.
- Technical Debt
- Technical debt is the future cost of shortcuts taken today: quick fixes, skipped tests, and messy structure that make the next change slower and riskier. A little of it is a normal, even smart trade-off to ship faster, but left unpaid it compounds until every new feature is a fight. AI-generated code can pile up debt fast when it is not reviewed, which is why I treat code review and clean conventions as first-class work, not an afterthought.
- Token
- A token is the small unit of text an LLM actually reads and writes, roughly a word or a piece of one, so "engineering" might be a couple of tokens. Models measure everything in tokens: the context window is a token budget, and you are billed per token in and per token out. Once you think in tokens, cost and the limits of a model stop being mysterious and become something you can plan around.
- Vibe Coding
- Vibe coding is building software by describing what you want to an AI and accepting most of what it produces, without reading the code closely. It is genuinely great for prototypes, demos, and throwaway experiments where speed is everything and the stakes are low. The trap is shipping vibe-coded work to production, where the accumulated bugs and technical debt eventually stall the project. A lot of my rescue work is turning a promising vibe-coded prototype into something that can actually scale.
Still sounds like a foreign language?
Bring me the term, the tool, or the half-formed idea. Book a call and I will translate it into plain English and tell you whether it is worth your time.