The Cold Email: Why It Still Works (and How AI Makes It Better)
Every week, I send about 50 cold emails. Not spam — targeted, researched, and individualized messages to decision-makers in SaaS, AI, and e‑commerce. My reply rate averages 18–22%, which is 4x higher than the industry norm. The secret? I don't write them. I vibe code them.
Vibe coding is the practice of using generative AI (like GPT‑4, Claude, or Gemini) to draft and iterate content while you curate the creative direction. Applied to cold email, it turns a painful, time‑intensive task into a scalable system that still feels human.
My Setup: From Research to Inbox
Before AI, I’d spend 30 minutes per email: stalking LinkedIn, reading their blog, and trying to craft a unique hook. Now I spend 3 minutes per email and get better results. Here’s the exact process:
- Gather raw intel – I copy a prospect’s LinkedIn bio, recent post, or company news into a text file.
- Feed the AI – I prompt: "You are a founder reaching out to [prospect name]. Write a 3‑sentence cold email that mentions [specific detail from their content]. Keep tone friendly, not salesy. End with a low‑friction ask (5‑min call)."
- Curate, don’t correct – I generate 3 variants, pick the best one, and tweak one or two words.
- Send with tracking – I use a plain‑text email via Mailgun, with a tracking pixel to know when they open.
This isn’t theory. This is how I’ve booked calls with C‑level execs at companies like Zapier and Shopify.
The ‘Vibe Coding’ Twist: Why It Wins
Traditional cold email advice says: keep it short, personalize, provide value. That’s still true. But AI lets you personalise at a depth that was impossible manually. For example:
- Hook based on a specific quote – I once saw a CEO say "We don’t do cold outreach" in an interview. I used that exact line as my opener. AI helped me rephrase it into a respectful challenge.
- Tailor pain points – If a prospect recently posted about hiring struggles, my email addresses that. AI drafts the paragraph in seconds.
- A/B test messaging – I generate 5 versions of the same email, each with a different framing (curiosity, authority, empathy). I send each to a small batch and let data decide.
| Approach | Time per email | Reply rate | Scalable? |
|---|---|---|---|
| Manual personalisation | 30 min | 10–15% | ❌ |
| AI‑assisted (vibe coding) | 3 min | 18–22% | ✅ |
| Blast / template | 0.5 min | 1–3% | ✅ |
Real Results: Numbers That Matter
To be concrete, here’s what I’ve tracked over the last 6 months (January – July 2026):
- Total cold emails sent: 1,200
- Average open rate: 68% (industry average for B2B cold email hovers around 20–30%)
- Average reply rate: 19.5%
- Calls booked: 47 (3.9% conversion from sent to meeting)
- Revenue directly attributed: ~$65,000 in new partnerships and contracts
One particular campaign stands out. I targeted 25 founders of AI startups. Instead of writing separate emails, I gave my AI assistant a folder with each founder’s latest Product Hunt launch. The assistant wrote 25 personalised emails in 10 minutes. 18 people replied. 5 turned into paying customers.
The Tool Stack I Rely On
- Email sending: I use Mailgun (SMTP) for deliverability, but for those learning the ropes, ASI Biont supports integration with Mailgun via API — see asibiont.com/courses for how to set it up.
- AI model: Mostly Claude 3.5 Sonnet for its long‑context understanding, and occasionally GPT‑4o for more creative hooks.
- Sequencing: A simple Python script sends follow‑ups after 3 days if no reply.
Common Pitfalls and How to Avoid Them
Vibe coding isn’t magic. I’ve made mistakes:
- Over‑personalising the wrong thing – AI can add irrelevant fluff. I once got an email back that said “Interesting that you mentioned my dog. He’s not part of my business.” Ouch. Now I only let AI use professional or public achievements, not personal guesses.
- Ignoring the ‘vibe’ – AI defaults to polite, generic corporate language. I always add a line like “Make it sound like a real person, slightly casual, with curiosity.”
- Forgetting to clean the list – Bounces hurt your domain reputation. I verify every email with a tool like NeverBounce before sending.
Conclusion: The Future Is Curation, Not Creation
Cold email isn’t dead. The spammy, copy‑paste version is. But when you combine genuine research with AI’s ability to reframe and refine, you get a tool that’s faster and more effective than either human or machine alone. That’s vibe coding.
If you’re still writing cold emails from scratch, start pairing yourself with an AI. Let the machine draft; you provide the context, the tone, and the final judgment. The results — higher replies, more meetings, less grunt work — speak for themselves.
Comments