Glossary

Agents and tools

AI agent

A model that decides its next step itself, calls tools and repeats until the task is done.

Agentic loop

The cycle of model call, tool call and result that an agent repeats until it stops.

Tool use (function calling)

The model returns a structured request to run a function you defined; your code runs it.

MCP (Model Context Protocol)

An open protocol for plugging tools and data into AI apps: write a server once, use it from any client.

Agent Skills

Folders with instructions, scripts and references that an agent loads only when the task needs them.

Subagent

A separate agent run with its own fresh context, started by a main agent for one subtask.

Harness

Everything around the model that makes it an agent: the loop, tools, permissions, memory, context management.

Human in the loop

A person approves, edits or rejects the agent’s action before it takes effect.

Context and prompts

Context window

The maximum number of tokens the model can read and write in one request.

Context rot

Model quality drops as the context fills with old, irrelevant or conflicting material.

Context engineering

Deciding what goes into the model’s context, in what order and when, across a whole task.

Compaction

Replacing the older part of a long conversation with a summary so the agent can keep going.

RAG (retrieval-augmented generation)

Finding relevant documents first and putting them into the prompt, so the model answers from them.

Embeddings

Vectors of numbers that place texts with similar meaning close together.

Structured outputs

The model’s answer is constrained to match a JSON schema you provide.

Quality and safety

Evals

A repeatable test set that scores an AI feature, so changes can be measured instead of guessed.

LLM as judge

Using a model with a rubric to grade another model’s output.

Hallucination

A fluent, confident answer that is false or not supported by the sources.

Prompt injection

Instructions hidden in data (a web page, an email, a file) that the model follows as if they came from you.

Guardrails

Checks around a model that block, fix or flag inputs and outputs that break the rules.

Slop (AI slop, workslop)

Low-quality content mass-produced with AI: it looks finished and says little.

Production and cost

Tokens

The pieces of text models read and write, and the unit you pay for.

Prompt caching

Reusing an identical prompt prefix across requests so it is cheaper and faster to process.

Cost cap

A hard spending limit enforced by your code, per user, per day or per month.

Model routing and fallback

Sending each request to the model that fits it, and to another one when the first fails.

AI gateway

A proxy between your app and model providers that adds logging, caching, limits, billing and fallbacks.

LLM observability

Tracing every model call with inputs, outputs, latency, tokens, cost and the user outcome.

Reasoning and effort

Models that think before answering, with a setting for how much thinking to spend.

Coding with agents

AGENTS.md and CLAUDE.md

Files in a repository with the rules a coding agent reads at the start of every session.

Coding agent

An agent that reads a repository, edits files, runs commands and tests, and opens pull requests.

Hooks

Scripts the harness runs automatically at fixed points, such as before a tool call or after an edit.

Spec-driven development

Writing the specification and acceptance tests first, then letting agents implement against them.

Vibe coding

Building by prompting and accepting the result without reading the code closely.

Software projects and pricing

MVP (minimum viable product)

The smallest version of a product that tests whether people want it enough to use or pay.

Riskiest assumption

The one belief that, if false, ends the idea. Test it first.

Fixed price

A contract where scope and price are agreed up front and the vendor carries the risk.

Time and materials (T&M)

Paying for the hours actually worked; scope can change, cost and timeline move with it.

Cone of uncertainty

How wide an honest estimate is at each stage: from 0.25× to 4× at the idea stage, narrowing as decisions are made.

Reference class forecasting

Estimating from what similar finished projects actually cost, instead of from your own feature list.

Scope creep

Small additions that each seem reasonable and together blow the budget and the date.

Email authentication

SPF (Sender Policy Framework)

A DNS record listing which servers may send email for your domain.

DKIM (DomainKeys Identified Mail)

A signature on each message, checked against a public key in your DNS.

DMARC

The policy that tells receivers what to do when mail fails SPF and DKIM, and where to report.

BIMI

Your logo next to your messages in supporting inboxes, once DMARC is enforced.

VMC and CMC

Mark certificates that prove you own the logo BIMI shows: Verified (trademark) or Common (12 months of use).

MTA-STS and TLS-RPT

A policy that makes senders deliver to you only over verified encryption, plus reports when it fails.