A single generalist assistant that finds prospects, judges them, and writes to them tends to do all three at a mediocre level — and you can't tune one step without disturbing the rest. A team splits the pipeline into narrow jobs: in Brohns' lead-generation flow, a Finder sources local businesses matching a niche and area (free, via OpenStreetMap), a Qualifier reads each website and produces an explainable 0–100 outdated score with a written reason, and an Outreacher drafts a first message that opens with a real, specific finding from that scoring. Each stage is gated on the one before it — the Outreacher cannot write until qualified leads actually exist. That gating is the practical difference between an agent team and a bundle of prompts.
Teams need coordination, and that's the orchestrator's job. In Brohns, the orchestrator is Bro: you describe a goal in plain language, Bro asks a few clarifying questions, then proposes a tailored team — usually two to seven agents, each with exactly one responsibility. The team lives on a canvas where every agent is a visible island, so you can watch what each one is doing and read its real reasoning on a timeline rather than a templated status message. Because responsibilities are narrow, you can teach, retune, or swap a single agent without rebuilding the whole workflow.
The pipeline deliberately stops one step short of the send button. Finding, scoring, and drafting run on their own; anything outward-facing — an email, an SMS, a spend — lands in a shared Approvals queue and waits for you. In Brohns that boundary is enforced on the server, with the recipient and content read from the database and guardrails like a daily send limit and a do-not-contact list applied, so no agent on the team can improvise its way past it. As the team earns your trust, you can move up the autonomy ladder and grant routine steps more independence — gradually, and always reversibly.