Creating a Sales Outreach Agent

Cold outreach is one of the most time-consuming and context-dependent sales tasks. The Cosine CLI can dramatically accelerate it by building an agent that knows your product deeply, understands what makes a good lead, and can draft personalised emails at scale.

The Setup

Create a dedicated outreach folder:

mkdir ~/sales-outreach
cd ~/sales-outreach
cos init

Then edit agent.md to define a sales persona with deep product context:

You are a world-class enterprise sales professional. You are precise, persuasive, and deeply customer-centric — never pushy, always relevant.

Your product:
- Cosine is an AI software engineering agent that multiplies developer productivity by 10–50x
- It works autonomously on code tasks, submits pull requests, and integrates with existing tools
- Key differentiators: Swarm Mode (proprietary parallel execution), 306 built-in LSP servers, persistent agent context via the file system

Your approach:
- Always research the prospect before drafting
- Lead with their specific pain point, not our features
- Keep emails under 150 words
- No generic openers ("Hope this finds you well", "I wanted to reach out")
- Reference something specific about their company or role

Data sources:
- Use the Attio MCP to find recently added customers
- Filter for contacts with a verified business email address
- Sort by ARR potential (highest first)

Running an Outreach Campaign

With the persona set up, give the agent a task:

“Find all customers added in the last 30 days. Discard anyone with a Gmail or invalid email. Sort the remaining 150 by ARR potential and draft a personalised outreach email for each. Save the drafts to outreach-drafts/.”

Switch to Swarm Mode for parallel execution — the agent will spawn subagents to research and draft for multiple prospects simultaneously, rather than processing them one by one.

Feeding in Product Context

The more context you give the agent about your product, the better the emails will be. You can add files directly to the folder:

• Vision doc: Drop in your product vision or positioning doc as vision.md.
• Case studies: Add customer case studies as individual files.
• Competitor analysis: Add a file comparing your product to alternatives.

Then reference them in your agent.md:

Before drafting any email, read `vision.md` to ensure accurate product positioning.
For enterprise prospects, also reference `enterprise-case-studies.md` for relevant examples.

Iterating on the Drafts

After the agent produces a batch of drafts, review a sample. If the tone is off or the emails are too long, refine the instructions in agent.md and re-run.

Because the drafts are saved as files in the folder, you can also:

• Ask the agent to revise specific drafts: “The email in outreach-drafts/acme.md is too feature-focused. Rewrite it to lead with their pain point.”
• Generate variations: “Give me three versions of the Acme email with different openers.”

The Compounding Effect

Each time you use this folder, the agent builds more context:

• It learns which email structures you prefer.
• It can reference past successful campaigns.
• It knows the objections you’ve heard and can preempt them.

Over time, the first draft it produces will require less revision.

Key Takeaways

• An agent.md with a strong sales persona and deep product context produces dramatically better outreach than a blank session.
• Use Swarm Mode to draft emails for many prospects in parallel.
• Feed in vision docs, case studies, and competitor analysis as context files.
• Iterate on agent.md based on what you find yourself correcting repeatedly.

---

Next: Sharing Context Between Agents