Build an outreach agent with the compliance gate inside it

An outreach agent that drafts and sends is a liability engine unless the compliance gate is built INTO it, not bolted on after. The pattern: the agent drafts channel-native messages (a text reads differently than an email), but every draft passes through a gate before it can go anywhere, and the gate can only ever return NEEDS-REVIEW, never 'cleared to send.' The gate checks the things that get people sued: consent for the channel, do-not-contact and DNC status, allowed contact hours, identification and a working opt-out, and whether the number or address is on a suppression or litigator list. The agent's job stops at a queued, gate-checked draft with its compliance flags attached; a human (you or counsel) makes the go decision. Two rules are absolute. The agent never auto-sends, it stages. And the gate never clears anything, it surfaces what's missing and hands off, because whether a specific send is lawful is a legal judgment that varies by channel and state and is not the model's call. Build the gate as code the draft must pass, and you get scale without handing a plaintiff's attorney your calendar.