glossary
the words of vibe coding, short and in plain language. whenever you click a dotted-underlined term on the site, you land here. no need to learn these upfront: look one up when you run into it.
- coding agent
- An AI program that doesn't just answer, it works: it turns your written request into code, creates and edits files, runs commands, and checks its own output. Claude Code and Codex are examples. You describe what you want in plain language; the agent carries it out.
- vibe coding
- Building software where you don't type the code: you describe what you want, and a coding agent writes it. Andrej Karpathy coined the term in 2025. The range goes from quick prototypes to production products; this site teaches the disciplined, production end of it.
- agentic engineering
- The mature, engineering-grade version of vibe coding: not "whatever comes out is fine", but a planned process where you direct the agents, verify the results, and every step is traceable. Agentic, because you don't type the code; engineering, because it's a craft you can learn.
- orchestration
- The work you do when building with agents: deciding what needs to happen, handing out the tasks (sometimes to several agents in parallel), and checking what comes back. Like a conductor: you don't play every instrument, but without you there is no music.
- build
- What we call each finished product: a build is a working, shipped piece of software made with agents. Every build comes with its full documentation (the recipe), so you see not just the result but how it was made.
- recipe
- The complete, step-by-step record of one build: the initial plan, the tasks the agents got, the cost, the time, and what went wrong. It's a recipe because it's repeatable: read it, then do the same on your own project.
- token
- AI models process text in small pieces (roughly word fragments) called tokens. Model usage is measured and billed in tokens, so the token cost tells you, in real money, what it cost to have an agent do a task.
- spec
- Short for specification: the written plan the agent works from. It says what to build, under which rules, and when it counts as done. The better the spec, the less you fix afterwards; writing a good spec is one of the most important parts of orchestration.
- workflow
- The schedule of a build: which agents work, in what order (one after another or in parallel), and who verifies what. Every recipe documents its workflow, because that's the part you can reuse on your own project.
- skill
- A pre-written guide the agent loads before working: rules, patterns and examples for one type of task (say, typography, animation or copywriting). Like a handbook the agent reads before starting.
- CLAUDE.md
- A plain text file in the project that the coding agent reads at the start of every session: it holds the project's rules, technical decisions and no-gos. It's the project's constitution: the agent doesn't deviate from it, so its quality shows directly in the results.
- session
- One continuous working session with the agent: the whole exchange from your request to the finished result, including every step the agent took. Our lessons replay real sessions, so you don't read a description, you watch the work happen.