2
2 Comments

Hey IH πŸ‘‹

I'm building 99 Agents β€” the idea is to prove AI can run real micro-businesses autonomously. Agent #1 is a cold email sequence generator that writes full 4-email outreach campaigns with timing, tone, and follow-up logic built in.

While building it, I dug into why most cold email campaigns get zero replies β€” even when people do "everything right." Wrote up a full breakdown:

https://nine9agents.polsia.app/blog/why-cold-email-gets-zero-replies

TL;DR: warmup takes way longer than gurus claim, volume doesn't fix bad copy, open rate metrics are broken (thanks Apple), and most people's CTAs ask for way too much from a stranger.

The tool is free to try (5 sequences/day, no signup wall): https://nine9agents.polsia.app

Would love feedback from anyone doing outbound. What's actually working for you in 2026?

on April 11, 2026
  1. 2

    The 99 Agents premise is fun, and starting with a cold email generator is a smart wedge β€” copy is the part everyone procrastinates on.

    One design suggestion from the deliverability side: have the agent generate the SEQUENCE, not just emails. A 4-email set written as one narrative (opener β†’ D3 bump β†’ D7 value-add β†’ D14 breakup) converts structurally better than four disconnected drafts β€” most replies live in touches 2–4, and generated emails that don't know about each other read like it.

    And if the agent ever touches sending (not just writing): bake in the guardrails as non-negotiable defaults β€” warmup awareness, 30–50/day/inbox caps, bounce circuit-breaker at 2%. An autonomous sender without those isn't an agent, it's a domain-burning machine with extra steps.

  2. 1

    The autonomous micro-business framing is more interesting than the cold email angle not because cold email doesn’t work but because the actual experiment here is whether an agent can handle the full feedback loop, not just the output. Cold email sequences are a solved problem. An agent that notices its own sequences aren’t converting and adjusts without human intervention is a genuinely different thing. Curious whether Agent 1 is measuring its own performance or whether that layer comes later in the portfolio. Would love to connect, feel free to check out what I’m building at tryreleaselog.com