The word "agent" earns its keep through a loop: take an action, observe what came back, decide what to do next. A prompt-and-response model answers whatever you type and stops; an agent keeps going until the goal is met or it genuinely needs you. That loop is what lets one piece of software find local businesses, read each website, score it, and draft a first message — four different kinds of work, chained together by its own judgment. In Brohns, the Qualifier agent makes the judging step explicit: it reads a lead's site, assigns an explainable 0–100 outdated score, and writes down why that lead is or isn't worth pursuing.
In practice, agents work best with one sharp responsibility each. An agent asked to find leads, qualify them, write outreach, and answer replies tends to do all four halfway; a Finder, a Qualifier, and an Outreacher each doing one job do it well. That's why Brohns doesn't hand you a single do-everything agent: you describe a goal in plain language and Bro, the orchestrator, assembles a team — usually two to seven agents, each with a single job. The team lives on a canvas where every agent is visible, including its live reasoning shown on a timeline as it works, so you're never guessing what software is doing on your behalf.
Because agents act instead of merely answering, the most important design question is where they stop. A Brohns agent researches, scores, and drafts entirely on its own, but the moment its work would touch the outside world — an outreach email, a payment reminder, anything sent or spent — it parks the item in an approvals queue and waits for your sign-off, a boundary enforced on the server rather than just drawn in the interface. Guardrails like a daily send limit, a send window, and a do-not-contact list sit underneath that. As an agent proves itself, you can loosen the leash step by step — and tighten it again whenever you want.