Why 5 agents, not one.
"AI SDR" packages too many jobs into one black box: it picks accounts, writes the email, sends, follows up, scores. When it fails — and it will — you can't tell which job broke. Five smaller agents, each with one job, fail visibly. You can swap any of them without rewriting the system.
The 5 roles.
- Signal Agent — watches funding, hiring, product usage, competitor, job-change sources 24/7. Output: a ranked list of "something just happened" rows.
- Research Agent — for each ranked account, produces a 1-pager: what they do, who decides, why now. Output: an SDR-ready brief.
- Content Agent — turns founder voice memos and decks into LinkedIn / X / newsletter / blog drafts. Output: a queue of drafts.
- Outbound Agent — composes context-driven emails and LinkedIn messages from the Research brief plus the Signal trigger. Output: drafts, never auto-send.
- CRM Agent — dedupes, validates stages, summarizes deals from email threads. Output: a clean weekly pipeline view.
What each playbook covers.
Role definition, the system prompt, tool integrations (Clay / Apollo / Notion / HubSpot / Slack), hand-off contract to the next agent, evals to run weekly, and the top 3 failure modes you have to design around. Plus a working example.
The principle.
Agents replace the boring 80% of GTM work. Humans own the judgment 20% — what to send, when to escalate, what's on-brand. The system collapses if you flip the ratio.
The Signal Agent maps to Buyer Signal Guides. The Outbound Agent maps to WF-01. The CRM Agent maps to WF-06.