The 42-Agent AI Sales Org. Here Are the Four You Start With.

Most people who hear "AI sales agent" picture one bot writing cold emails. That is a feature, not a sales team.

A sales team has an org chart. Someone prospects. Someone qualifies. Someone preps the call. Someone writes the follow-up. Someone keeps the CRM clean so none of it falls apart. When you replace that team with AI, you are not replacing one job. You are rebuilding the whole chart.

This is the full chart. Forty-two agents, five teams, each with a ready-to-paste prompt and a tool stack. And the honest part almost nobody says out loud: you do not deploy forty-two agents. You deploy four, prove they earn their keep, then add the next four. The businesses that win with this are not the ones running the most agents. They are the ones running the right agents in the right order.

What the 42 Agents Actually Do

Five teams. Each owns one stage of the revenue motion. Here is the whole map before we go deep on the four that matter first.

The point of the chart is not the headcount. It is the separation of duties. Each agent has one job, one prompt, one set of success metrics. That is what makes the system debuggable. When reply rates drop, you know which agent to fix. When the CRM fills with duplicates, you know which agent failed. A single mega-prompt trying to do all five stages is a black box. Forty-two small prompts is a machine you can read.

There is an interactive version of the org chart you can click through, team by team, agent by agent: the full 42-agent map. No login. Send it to whoever needs to see the whole picture.

Now the four you build first.

Agent One: The Orchestrator

Start at the top. Not because it is glamorous, but because without it every other agent works in isolation and nobody coordinates the handoffs.

The Chief Sales Officer agent is a router. A lead comes in, it decides which team handles it. A deal gets big, it flags a human. It does not write emails or score leads. It reads the situation and sends the work to the right place with the context attached.

Deploy this first with just two teams wired underneath it. Add the rest as you build them.

Copy: the orchestrator prompt

You are the Chief Sales Officer Agent, the central orchestrator of an AI sales
organization.

CORE FUNCTIONS:
1. INTAKE & TRIAGE: Determine which team and manager should handle each request.
2. DELEGATION: Route tasks with clear context and priority level (P1/P2/P3).
3. MONITORING: Track task completion, flag delays, identify anomalies.
4. ESCALATION: Escalate when deal value exceeds threshold, legal or compliance
   questions arise, the customer requests human contact, or a strategic decision
   is needed.
5. REPORTING: Synthesize cross-team data into actionable insights.

TEAMS YOU OVERSEE:
- Channel Outreach Manager -> Outreach Team
- Global BDR Manager -> Multilingual BDR Team
- Systems & Data Manager -> RevOps Team
- Competitive Intel Manager -> Intel Team
- Qualification Manager -> Research Team
- Call Intelligence Manager -> Call Intel Team
- Workspace Manager -> Logistics Team
- Content Enablement Manager -> Content Team

When delegating, always include: Task summary, Priority, Deadline, Required
context, and Success criteria.

Your first real run: give it one incoming lead and tell it to route the lead. Watch what it decides. If it routes a "pricing question" to Outreach instead of to your Product agent, your team definitions are wrong, not the model. Fix the routing rules in the prompt before you trust it with volume.

The mistake that makes it fail: wiring it to escalate everything. An orchestrator that flags every deal for a human is just a slower inbox. Set a real dollar threshold. Below it, the machine runs. Above it, you look.

Agent Two: The Email Specialist

This is the agent everyone wants first. Deploy it second, after the orchestrator can route its replies.

The whole prompt is built around one rule: write like a human who did their homework, not a template that found a merge field.

Copy: the email specialist prompt

You are the Email Outreach Specialist.

SUBJECT LINES: 4 to 7 words, lowercase, no clickbait, curiosity-driven.
Examples: "quick question about [company]", "[mutual connection] mentioned you"

EMAIL STRUCTURE:
- Opening: Personalized observation. Never "I saw your LinkedIn".
- Bridge: Connect the observation to their likely problem.
- Value: One clear, specific benefit.
- CTA: Single, low-friction ask.
- Length: 50 to 100 words maximum.

PERSONALIZATION LEVELS:
- L1: name and company
- L2: recent news or changes
- L3: specific challenges or tech stack
Always aim for L2 or higher.

FOLLOW-UP: Day 3 new value, Day 6 different angle, Day 10 breakup email.
Never more than 4 emails without engagement.

REPLY HANDLING:
- Positive -> hand to the Meeting Scheduler
- Questions -> answer, then CTA
- Objections -> empathy, then an alternative
- Not interested -> thank them, leave a future door open

TONE: Conversational, peer-to-peer, zero fluff.

Your first real run: feed it one real prospect with a real trigger event, funding, a job posting, a product launch. Read the draft it produces. If the opening line could be sent to a hundred people unchanged, it hit L1 and stopped. Reject it. The whole value of this agent is the second sentence knowing something specific.

The mistake that makes it fail: letting it send. Deliverability dies when a fresh agent blasts volume. Cap it at 30 sends per account per day, warm the inboxes first, and verify addresses before the first send. That is not a nice-to-have. A 5 percent bounce rate on cold volume gets the whole domain flagged.

Agent Three: The Lead Scorer

Outreach without scoring means your best-fit leads sit in the same queue as tire-kickers. This agent puts a number on every lead so the machine knows who to chase first.

It is a scoring model, not a judgment call. Firmographic fit, behavioral signals, engagement, minus the negatives. The output is one number and a bucket.

Copy: the lead-scoring prompt

You are the Lead-Scoring Specialist.

FIRMOGRAPHIC (0-40):
- Company size (ideal band scores highest)
- Industry (Tier 1 verticals score highest)
- Geography (primary markets score highest)

BEHAVIORAL (0-40):
- Pricing page visit = 10
- Demo page visit = 8
- Email reply = 15
- Email click = 8
- Whitepaper download = 7

NEGATIVE:
- Unsubscribe = -20
- Hard bounce = -10
- Competitor = -40
- Clear bad fit = -20

THRESHOLDS:
- Hot (80+): priority outreach
- Warm (50-79): active sequence
- Cool (25-49): nurture
- Cold (under 25): low priority

OPERATIONS: score behavioral signals in real time, recalculate daily,
review thresholds weekly, review the whole model monthly.

Your first real run: score fifty existing leads and sort them. Then look at who actually became a customer last quarter. If your closed deals are landing in Cool and Cold, your weights are backwards. The model is only as good as the correlation between its score and your real outcomes.

The mistake that makes it fail: setting the numbers once and never touching them. A scoring model is a living thing. Markets shift, your ICP tightens, the signals that mattered last year stop mattering. Review it monthly or it slowly starts lying to you.

Agent Four: The Call Prep

Your reps, human or AI, walk into calls cold. This agent walks in for them. It researches the prospect and hands over a one-page brief two hours before the call, so nobody opens a conversation with "so, tell me about your business."

Copy: the call-prep prompt

You are the Discovery-Call Prep Specialist.

BUILD THIS PREP DOCUMENT:
1. COMPANY SNAPSHOT: name, industry, size, funding, locations
2. CONTACT INTEL: title, tenure, LinkedIn highlights, communication style
3. BUSINESS CONTEXT: recent news, job postings, tech stack, competitors
4. PAIN HYPOTHESES: 2 to 3 evidence-based guesses, each with a question to
   validate it
5. DISCOVERY QUESTIONS: Opening -> Situation -> Problem -> Impact -> Future
6. LIKELY OBJECTIONS: the ones you expect, with a response for each
7. TALKING POINTS: relevant case studies, features to raise, ROI numbers
8. RED FLAGS: any signal that this could stall or die

DELIVERY: one page, TL;DR at the top, in the rep's hands 2 hours before
the call.

Your first real run: pick your next real booked call and have it build the brief. Read the pain hypotheses. If they are generic ("they probably want to save time"), the research was shallow. A good hypothesis points at evidence: "they posted three ops-manager roles in two months, so they are drowning in manual work they cannot hire out of fast enough."

The mistake that makes it fail: treating the brief as a script. It is a map, not a monologue. The rep still has to listen. A brief that gets read out loud instead of used quietly is worse than no brief.

The Build Order

Four agents get you a working loop: route, reach out, score, prep. That is a real sales motion. Everything else in the chart is amplification on top of a system that already works.

Here is the sequence the playbook lays out, so you expand in the order that compounds.

• Weeks 1 to 2. Foundation. The orchestrator plus RevOps plumbing. Wire the CRM and email. Add the Monitoring agent so you can see what is happening.
• Weeks 3 to 4. Outreach. The Email Specialist and the Channel Manager that keeps your channels from hitting the same prospect twice in a day.
• Weeks 5 to 6. Intelligence. Qualification, scoring, prospect research, and call prep. This is where quality jumps.
• Weeks 7 to 8. Scale. Voice, video, events, the multilingual reps, and the full content team.

You do not have to run this on an eight-week clock. You have to run it in this order. Scaling outreach before you have scoring means you scale noise. Adding voice and video before the email loop converts means more channels producing the same weak result.

The Tech Stack

You do not need enterprise tooling to start. Here is the split between what gets you running and what you add when volume justifies the cost.

The "start with" column runs the four core agents. The prompts above are engine-agnostic. They work in Claude, in Codex, in any orchestration tool that can hold a system prompt. The model is not the moat. The org design is.

Honest: Why 42 Is the Wrong Number to Start At

The temptation is to build the whole chart because the whole chart is impressive. Do not.

Forty-two agents is forty-two things that can break, forty-two prompts to maintain, forty-two handoffs to debug. Stand all of that up on day one and you cannot tell which agent caused the mess. You get a system nobody trusts and everybody quietly stops using.

There are also jobs on this chart that should stay partly human for now. Objection handling on a real deal. Anything touching legal or compliance. Pricing conversations above your escalation threshold. The playbook is built to escalate exactly these, and that guardrail is the point, not a limitation. An AI sales org that never hands off is not more advanced. It is more dangerous.

And the multilingual reps, the eight-language coverage, only matter if you sell into those markets. Building agents for languages you have no prospects in is headcount theater. Build for the pipeline you actually have.

The measure of this system is not how many agents are live. It is whether the four core agents move a real number: meetings booked, pipeline created, hours you got back. If four agents do that, you earned the right to build the fifth.

If You Only Do One Thing

Deploy the orchestrator and the Email Specialist this week. Two agents. Wire the orchestrator to route replies to the Email Specialist and escalate anything above a dollar threshold you set.

Give it ten real prospects. Read every draft before it sends. You are not launching a forty-two-agent org. You are proving that two of them can do one loop without you. That is the whole test. If it holds, the other forty are just the same move, repeated.

If it does not hold, you spent an afternoon and learned exactly which rule to fix. That is the cheapest sales hire you will ever make.