Glossary

The real-time, filterable stream of every agent step — tool calls, edits, commits, messages — that Black Box shows the owner.

The pattern where one agent hands a scoped sub-task to another, with inputs, success criteria, and an expected return shape.

A dedicated agent (or rubric-guided model) that grades the output of other agents against acceptance criteria before it ships.

The coordination layer that assigns work, sequences dependencies, and aggregates results across multiple AI agents.

A reusable, domain-specific procedure — goals, steps, tools, checkpoints — that an agent follows to deliver a repeatable outcome.

AI that plans, takes actions with tools, observes results, and self-corrects toward a goal without step-by-step human instruction.

A software system built on a language model that perceives context, chooses actions from a toolset, and executes them in a loop to reach a goal.

AI automation uses models to make judgment calls inside a task; intelligent automation combines that with process mining, RPA, and workflow orchestration across a business.

A software product that behaves like a business — a CEO agent, specialist agents, shared memory, and the authority to act on the owner’s behalf.

A configured set of specialist agents — marketing, engineering, sales, design, operations — coordinated by a CEO agent to run a business function end to end.

A bundle of playbooks, prompts, tools, and evaluations that gives an AI company new domain-specific capabilities in one install.

A queue of high-stakes agent decisions — spend, sends, deploys — that the human owner explicitly approves or rejects before execution.

Astro is a static-site generator that ships zero JavaScript by default, ideal for fast, SEO-friendly marketing sites.

An AI agent that can complete a multi-step goal without human intervention between steps, deciding its own sub-tasks and tool calls.

The use of software to execute recurring, multi-step business processes — onboarding, invoicing, reporting — with minimal human touch.

An agent that reads calendar context, proposes meeting times, drafts invites, and resolves conflicts on the owner’s behalf.

The top-level orchestrator agent that decomposes a goal, delegates to specialists, tracks progress, and synthesizes the final result.

A prompting technique that asks the model to reason step by step, improving accuracy on multi-step problems.

A saved state — message log, scratchpad, tool output — that lets an agent resume or branch a long-running session without losing context.

A safeguard that halts an agent loop when it exceeds budget, loops on errors, or breaches a policy, preventing runaway behavior.

Anthropic’s official SDK for building agents on Claude, providing session management, tool use, subagents, and hooks out of the box.

Software that sources prospects, personalizes outbound messages, sends at scale, and routes replies into a CRM or inbox triage flow.

The repeatable flow from idea to published content — research, draft, edit, illustrate, publish, distribute — coordinated by specialist agents.

The maximum number of tokens a language model can attend to at once, including the prompt, conversation history, and tool outputs.

Google’s set of user-experience metrics — Largest Contentful Paint, Interaction to Next Paint, Cumulative Layout Shift — that factor into search ranking.

The use of AI and rules to enrich leads, update stages, draft follow-ups, and keep a CRM accurate without manual data entry.

Google’s quality framework — Experience, Expertise, Authoritativeness, Trust — that search raters use to evaluate content credibility.

A check in an agent workflow where an evaluator must pass the output against a rubric before downstream steps run or results reach the user.

The practice of optimizing content so that answer engines — ChatGPT, Claude, Perplexity, Gemini — cite and quote it in their responses.

A description of the firmographic, technographic, and behavioral traits of the customer most likely to buy, renew, and refer.

The automated sorting, drafting, and escalation of incoming email so the owner only sees messages that require a human decision.

A JSON-based format for embedding structured data — organization, product, article, FAQ — in web pages for search and answer engines.

A pipeline where agents discover prospects, enrich their data, score fit against an ICP, and hand qualified leads to sales.

The programmatic collection of contact and firmographic data on prospects from public sources, directories, and social platforms.

A proposed root-level text file that tells language models which pages of a site to prioritize, and how to cite them.

An OAuth pattern where the redirect URI is a local loopback address (127.0.0.1) on an ephemeral port, used by native and desktop apps.

A process that exposes tools, resources, and prompts to an MCP client over stdio, HTTP, or SSE following the Model Context Protocol spec.

A named, typed function exposed by an MCP server that an AI agent can call with structured arguments to read data or take an action.

An open protocol that lets AI applications connect to external tools and data sources through a uniform client-server interface.

A software system composed of multiple cooperating AI agents, each with a role, tools, and memory, coordinating to solve a shared goal.

Proof Key for Code Exchange — an OAuth 2.0 extension that protects public clients from authorization-code interception attacks.

An API feature that stores and reuses large, repeated prompt prefixes across requests to reduce latency and cost.

Railway is a managed cloud platform for deploying Node, Python, and Go services from GitHub with built-in Postgres, Redis, and domains.

A protocol pattern where a client invokes a named procedure on a server with typed arguments and receives a typed result.

A pattern where a language model retrieves relevant documents from a knowledge store and grounds its answer in that retrieved context.

RPA follows fixed scripts across GUIs; AI agents plan, adapt, and handle ambiguity using language models and tools.

A technique where an agent checks its own output against criteria — tests, schemas, rubrics — and retries if the check fails.

A one-way streaming protocol over HTTP that pushes text events from a server to a browser as they occur.

An OAuth 2.0 flow where the authorization code is exchanged for tokens on a trusted backend, keeping client secrets off the user’s device.

The use of AI and workflow software to deliver a professional service — research, content, design — end to end with minimal human hours.

An AI agent scoped to a specific role — coding, research, content, browsing — with role-shaped prompts, tools, and evaluation criteria.

Machine-readable metadata — usually JSON-LD — that describes page entities and enables rich results, sitelinks, and answer-engine citations.

A model response constrained to a schema — JSON, XML, or a grammar — so downstream code can parse it reliably.

A component that compresses an agent’s growing conversation history into a shorter summary so the session can continue past the context-window limit.

Supabase is an open-source backend platform built on Postgres, providing auth, storage, realtime subscriptions, and edge functions.

Tauri is a Rust-based framework for building lightweight, cross-platform desktop apps that render a web UI in a native webview.

The per-task ceiling on input and output tokens an agent may consume, enforced to control cost and latency.

The capability for a language model to call external functions — search, code execution, APIs — with structured arguments and use the results in its next response.

The use of software to execute a sequence of steps across apps based on triggers, conditions, and scheduled events.

A prompting style that asks a model to perform a task with no examples, relying on instructions and the model’s prior training.