Build a Marketing Agent with AutoGPT (Generate Campaigns)

# Clone and set up AutoGPT
git clone https://github.com/Significant-Gravitas/AutoGPT.git
cd AutoGPT/classic
pip install -r requirements.txt

# .env
OPENAI_API_KEY=sk-your-key-here
MEMORY_BACKEND=local
BROWSE_CHUNK_MAX_LENGTH=8000
BROWSE_SUMMARY_MAX_TOKEN=300

mkdir marketing_output

PRODUCT BRIEF
=============
Product Name: FlowDesk
Product Type: B2B SaaS project management tool
Target Audience: Remote engineering teams at 20-100 person startups
Key Value Prop: AI-powered sprint planning that reduces planning meetings by 60%
Price Point: $15/user/month, $12/user/month annual
Key Features:
  - AI sprint planner that estimates task complexity automatically
  - Async standup summaries replacing daily sync calls
  - GitHub/Linear/Jira integrations
  - Custom burndown reports with predictive completion dates
Competitors: Linear, Jira, Asana, Monday.com
Brand Voice: Direct, technical, no fluff. We speak to engineers, not managers.
Campaign Goal: Drive 500 trial signups in 30 days
Budget Context: Paid social + email to a list of 8,000 subscribers

BRAND VOICE EXAMPLES:
Example 1: "Your sprint planning meeting is a symptom. FlowDesk is the cure."
Example 2: "Stop estimating. Start knowing."
Example 3: "Built by engineers who hated the tools they had to use."

ai_name: MarketingAgent
ai_role: >
  A specialized B2B SaaS marketing agent. You read product briefs and create
  complete campaign packages. You write in the brand voice defined in the brief.
  You save all outputs to the marketing_output directory.

ai_goals:
  - Read product_brief.txt and understand the product, audience, and campaign goals
  - Research 4-6 competitors mentioned in the brief and create a comparison table saved to marketing_output/competitor_analysis.md
  - Create a 30-day campaign strategy document saved to marketing_output/campaign_strategy.md
  - Write a 5-email onboarding sequence for trial users saved to marketing_output/email_sequence.md
  - Create 15 social media posts for LinkedIn and Twitter saved to marketing_output/social_calendar.md
  - Write 3 Google Ad and 3 Meta Ad variations saved to marketing_output/ad_copy.md
  - Create a landing page outline saved to marketing_output/landing_page_outline.md

python -m autogpt --ai-settings ai_settings.yaml --continuous --continuous-limit 50

# FlowDesk Trial Onboarding Email Sequence

## Email 1 — Day 0: Welcome (sent immediately after signup)
**Subject:** Your FlowDesk trial starts now (here's what to do first)
**Preview:** Skip the setup video. Here's the 3-minute path to your first AI sprint.

You're in.

FlowDesk works differently than the tools you've used before. The AI doesn't
just help you organize tasks — it estimates complexity based on your team's
actual velocity, not guesswork.

Start here:
→ Connect GitHub (takes 60 seconds)
→ Import your backlog (CSV or paste)
→ Run your first AI sprint plan

Your first AI sprint plan is ready in under 5 minutes.

[Start your first sprint →]

— The FlowDesk Team

P.S. If you get stuck, reply to this email. We read every one.

---

## Email 2 — Day 3: Feature spotlight
**Subject:** The meeting you won't need to have anymore
**Preview:** FlowDesk's async standup feature is replacing daily syncs for 2,000+ teams.

...

# FlowDesk Ad Copy Variations

## Google Search Ads

### Variation A
Headline 1: AI Sprint Planning for Dev Teams
Headline 2: Cut Planning Meetings by 60%
Headline 3: Try Free for 14 Days
Description: FlowDesk auto-estimates task complexity from your commit history.
Built for engineers who hate meetings. No credit card required.

### Variation B
Headline 1: Replace Your Daily Standup
Headline 2: AI-Powered Project Management
Headline 3: Integrates with GitHub & Linear
Description: Async standup summaries. AI sprint plans. Predictive burndown charts.
Engineering teams at 500+ startups use FlowDesk.

### Variation C
Headline 1: Sprint Planning in 5 Minutes
Headline 2: FlowDesk for Remote Dev Teams
Headline 3: Start Free Today
Description: Stop spending 2 hours planning what you could plan in 5.
FlowDesk AI learns your team's velocity and estimates effort automatically.

ai_goals:
  - Read all files in marketing_output/
  - Review each for: brand voice consistency, factual accuracy, and call-to-action clarity
  - Rewrite any section scoring below 7/10 on these dimensions
  - Save a QA_report.md with your assessment of each piece

# post_process.py — Convert agent output to HubSpot-ready format
import json
import re
from pathlib import Path

def parse_email_sequence(markdown_path: str) -> list[dict]:
    """Parse the agent's email sequence markdown into structured email objects."""
    text = Path(markdown_path).read_text()
    emails = []
    
    # Split on email number headers
    sections = re.split(r"## Email \d+", text)[1:]
    
    for i, section in enumerate(sections, 1):
        # Extract subject and body
        subject_match = re.search(r"\*\*Subject:\*\* (.+)", section)
        preview_match = re.search(r"\*\*Preview:\*\* (.+)", section)
        
        # Get body (everything after preview line)
        body_start = section.find("\n\n", section.find("Preview:")) + 2
        body = section[body_start:].strip()
        
        emails.append({
            "email_number": i,
            "subject": subject_match.group(1) if subject_match else f"Email {i}",
            "preview_text": preview_match.group(1) if preview_match else "",
            "body": body,
            "send_delay_days": [0, 3, 7, 14, 30][i-1] if i <= 5 else (i-1) * 7,
        })
    
    return emails

emails = parse_email_sequence("marketing_output/email_sequence.md")
print(json.dumps(emails[:1], indent=2))