The vocabulary, without the mystique.

Agent canvas

An agent canvas is a visual workspace that shows every AI agent on a team as its own element: what it is working on right now, what it is reasoning through, and how work flows from one agent to the next. Instead of reading logs after the fact, you watch the team operate in a single live view. It turns an otherwise invisible multi-agent process into something you can inspect at a glance.

AI agent

An AI agent is software that pursues a goal by deciding on and carrying out multi-step actions — searching, judging, drafting — rather than only answering the prompt in front of it. Where a chat assistant responds and stops, an agent breaks a goal into steps, executes them, checks its own results, and comes back with finished work for you to review.

AI agent team

An AI agent team is a group of specialized AI agents that share one pipeline, with each agent owning a single, sharp responsibility — one finds, one qualifies, one drafts, one hands off for approval. Work passes down the line in sequence, so each step is gated on the previous one instead of a single assistant juggling everything at once. An orchestrator coordinates the agents toward a shared goal and decides which specialist the team needs.

AI credits

AI credits are a usage-based billing unit that meters the work AI agents actually perform: each discrete action — finding leads, qualifying them, drafting a message — deducts a small, fixed number of credits. Because credits track output rather than seats or time, a quiet month costs little and idle agents cost nothing.

Answer Engine Optimization (AEO)

Answer Engine Optimization (AEO) is the practice of structuring content so AI answer engines — ChatGPT, Perplexity, Google's AI Overviews — cite it in the answers they generate. Where traditional SEO competes for a position in a list of links, AEO competes to be the source an AI quotes when someone asks a question. It rewards pages that open with a direct answer, state verifiable specifics, and are easy for machines to parse.

Approval-first

Approval-first is a design principle for AI agents in which every outward-facing or irreversible action — sending an email, publishing a post, spending money — requires explicit human approval before it executes. Agents research, score, and draft autonomously, but the moment an action would touch the outside world, it stops and waits for a person to approve, edit, or dismiss it. The principle is strongest when enforced on the server, so no prompt injection or UI bug can slip past the approval step.

Autonomy ladder

An autonomy ladder is a graduated trust model for AI agents: they begin by proposing every action for human approval, earn permission to handle routine work independently within explicit limits, and only gradually gain broader autonomy. Every rung is granted deliberately by the user, and every rung is reversible.

Cold outreach

Cold outreach is a first-contact message — usually email — sent to someone who has no prior relationship with you, to open a conversation about your product or service. It differs from spam in three ways: the message is researched and specific to the recipient, it's sent in limited, deliberate volumes, and it makes opting out easy. Done well, cold outreach reads like one professional writing to another about something they actually noticed.

Do-not-contact list

A do-not-contact list is a suppression list of addresses that must never receive a message, no matter what a campaign or agent tries to send. Entries come from unsubscribes, hard bounces, spam complaints, and manual additions like existing clients or people who asked to be left alone. The list only works if it is enforced at send time — checked at the moment a message goes out, not just when it is drafted.

Dunning

Dunning is the systematic process of reminding customers to pay overdue invoices, using a sequence of messages that escalate in firmness as the invoice ages. A typical sequence moves from a friendly nudge just past the due date to firmer notices as weeks pass — always professional, because the recipient is a paying customer you want to keep. Done well, dunning recovers revenue without burning the relationship.

Generative Engine Optimization (GEO)

Generative Engine Optimization (GEO) is the practice of making your content easy for generative AI engines — ChatGPT, Perplexity, and the AI answers embedded in search — to retrieve, understand, and cite. It is a near-synonym of answer engine optimization (AEO): both describe the shift from ranking in link lists to being quoted inside AI-generated answers. The "generative" framing emphasizes how models compose answers from multiple sources; the tactics are the same either way.

Human-in-the-loop (HITL)

Human-in-the-loop (HITL) is a design pattern where an automated system pauses at defined checkpoints so a person can review, edit, or reject its output before the process continues. Unlike human-on-the-loop setups, where a person merely monitors automation and intervenes after the fact, HITL makes human judgment a required step: the system cannot act without it. In AI agent workflows, the checkpoint typically sits right before any outward-facing action, like sending an email or spending money.

Lead qualification

Lead qualification is the decision, made before any outreach, about whether a prospect is worth pursuing at all. Unlike a numeric score, a qualification verdict comes with a written reason: concrete evidence of a problem you can solve and a plausible fit. Disqualifying a lead counts as a win — it protects your reply rate, your sender reputation, and your time.

Lead scoring

Lead scoring is the practice of ranking prospects with a numeric value that reflects how well they fit your offer and how likely they are to respond. Instead of working a list top to bottom, you work it in order of expected payoff. A score is only as useful as it is explainable: you should be able to see exactly which signals produced the number.

llms.txt

llms.txt is an emerging web convention: a plain-Markdown file served at /llms.txt that gives AI crawlers a curated entry point to a site. It opens with an H1 naming the brand, a one-line description of what the site is, and annotated links to the pages a language model should read first. Unlike robots.txt, it blocks nothing — it points models toward your best content instead of away from anything.

Orchestrator

An orchestrator is the coordinating layer of an AI agent team: it assembles the right specialists for a goal, routes work between them, and gates each stage so output only moves forward when it's ready. It doesn't do the work itself — it decides who works on what, in what order, and which steps need human sign-off. In Brohns, the orchestrator is Bro, the assistant that turns a plain-language goal into a working team.

Send window

A send window is a configured time range — usually business hours — during which outbound messages like cold emails are allowed to go out. Attempt a send outside that range and the guardrail blocks it: you approve, or approve again, once the window is open. It's both a deliverability guardrail and a basic courtesy: mail that arrives during the workday gets read, while mail that arrives at 3 a.m. announces itself as automation.

Structured data

Structured data is machine-readable markup — most commonly schema.org vocabulary written as JSON-LD — added to a web page so search engines and AI answer engines can understand exactly what the page contains. Instead of leaving crawlers to infer meaning from prose, it labels the facts explicitly: this is an Organization, this is an FAQPage with these exact questions and answers, this is an Article by this author on this date. It is a core technical ingredient of both classic SEO and answer engine optimization.