Module 10 — Putting the AI Agent to Work

The previous nine modules describe a complete cold-pitching practice. Doing it by hand takes between 30 minutes and 2 hours per prospect — research, drafting, scheduling, follow-up tracking, deliverability checking, A/B test bookkeeping. That’s the difference between “I’ll cold-pitch when I have time” and “I cold-pitch 15 prospects a week, every week, without thinking about it.”

This module is about the AI agent — Cold Pitch Boss — that automates the mechanical 80 percent of the work so you can focus on the 20 percent that’s actually your judgement: picking targets, approving drafts, and showing up on the calls.

What it does

Given an Ideal Client Profile and a list of seed prospects, Cold Pitch Boss:

1. Researches each prospect across Google, the company’s website, their blog, their press releases, LinkedIn, and the Facebook Ads Library. Returns the five research topics from module 5: revenue signal, recent trigger event, vertical, company culture, individual prospect notes.
2. Scores prospects against your ICP on the 10 attributes from module 5. Skips anyone scoring below 7/10. Skips anyone whose most recent trigger event is older than 90 days.
3. Composes a pitch using the PAS framework and the 8-part anatomy from module 6. Personalized to the research it just did. No “I hope this email finds you well.” No spam-trigger words. Plain text.
4. Lints the draft against the 10 principles from module 4 plus the deliverability rules from module 7. Auto-rewrites up to two times if anything fails. Flags anything still failing for human review.
5. Creates a Gmail draft for you to review. (Drafts-first by default. You explicitly pass --send to make it send.)
6. Schedules follow-ups at the cadence from module 8 — Tuesday 10am → Thursday 8am → next Tuesday 1pm — in the prospect’s timezone, not yours.
7. Tracks everything — opens, replies, link clicks, unsubscribes, spam reports, booked calls — in a local SQLite database so you can see what’s working.

What it does NOT do

It does not eliminate your judgement. It does not magically find prospects you didn’t tell it about. It does not write better pitches than you would on your best day — it writes the pitch you would write on your best day, every single time, in 30 seconds.

It also does not auto-send anything unless you explicitly tell it to. The default mode creates Gmail drafts and you review them. This is intentional. The first 50 pitches the agent generates for you should be reviewed line-by-line so you build confidence in the output. Past that, you can switch to send-mode for prospects that meet a higher ICP threshold.

Installing it

The agent is a Python project. It runs on any machine — Mac, Linux, Windows via WSL — that has Python 3.10+.

The fastest path is the Install with Claude flow: download the zip from the download page, paste one prompt into Claude (Code or chat), and Claude walks you through every step. About 30 minutes start-to-finish, mostly answering questions about your business.

If you’d rather do it by hand:

# 1. Download cold-pitch-boss.zip from freelance.mikee.ai/download
unzip ~/Downloads/cold-pitch-boss.zip -d ~/cold-pitch-boss
cd ~/cold-pitch-boss

# 2. Create a virtual environment and install dependencies
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
python -m playwright install chromium

# 3. Set your Anthropic API key (this is what writes the pitches)
export ANTHROPIC_API_KEY=sk-ant-...

Setting up Gmail (one-time)

1. Create a Google Cloud project. Enable the Gmail API.
2. Configure the OAuth consent screen (External; add your email as a test user).
3. Create OAuth credentials, type “Desktop.” Download the JSON.
4. Save it as cold_pitch_agent/data/gmail/credentials.json.
5. First run will pop a browser for consent and save a token.json next to it.

Writing your ICP

Copy cold_pitch_agent/icp_example.yaml to my_icp.yaml and fill it in based on module 5’s 10 attributes. The example file walks through every field.

The single most important field is root_problem — the specific problem you solve for clients. Be honest and concrete. I help Series B B2B SaaS companies that have lost editorial cadence after a marketing-team turnover. If you can’t write this sentence yet, your ICP isn’t finished.

Running your first campaign

The seed prospects file is a JSON list — one entry per company you want to pitch. The agent will research each one and apply your ICP filter.

python -m cold_pitch_agent.agent \
  --icp my_icp.yaml \
  --prospects my_prospects.json \
  --max 5

This will:

• Research the 5 prospects
• Skip any that don’t match your ICP or that lack a recent trigger event
• Compose drafts for the ones that pass
• Save the drafts to your Gmail drafts folder
• Schedule the follow-ups

Open your Gmail drafts. Review each pitch. Edit as needed. Send when ready.

When you want the agent to send pitch #1 automatically (still queueing the follow-ups for your review), add --send:

python -m cold_pitch_agent.agent \
  --icp my_icp.yaml \
  --prospects my_prospects.json \
  --max 15 \
  --send

Reviewing the results

Everything goes into cold_pitch_agent/data/cold_pitch.db — a small SQLite file. Open it with any SQL tool. The schema is in cold_pitch_agent/storage/schema.sql.

Useful queries:

-- Booked-call rate
SELECT COUNT(*) FILTER (WHERE kind = 'booked_call') * 1.0 / COUNT(DISTINCT prospect_id) AS rate
FROM events;

-- Open rate by subject-line variant (A/B testing)
SELECT ab_variant, COUNT(*) FILTER (WHERE e.kind = 'open') * 1.0 / COUNT(*) AS open_rate
FROM pitches p JOIN events e ON e.pitch_id = p.id
GROUP BY ab_variant;

-- Prospects on warm bench
SELECT * FROM prospects WHERE status = 'warm_bench';

The first month

In your first month with the agent:

Week 1: Build a 30-prospect seed list. Run the agent in draft-only mode. Review every draft. Edit anything that doesn’t sound like you. Take notes on what the agent gets wrong; this is feedback for tuning the system prompt later.

Week 2: Send your first batch of pitches (still reviewing each draft, but now actually clicking send in Gmail). Watch the open rates. Don’t expect replies yet.

Week 3: Follow-ups #2 and #3 start hitting. Replies start to come in. Track booked-call rate.

Week 4: Tune your ICP. Drop the attributes that aren’t producing results. Tighten the ones that are. Add 30 more prospects to the seed list.

By the end of month one you’ll have a calibrated process, a tuned ICP, and a few booked calls — and you’ll be ready to ramp to 15 prospects a week as your steady-state rhythm.

That’s the whole job.