Last week I got a sales pitch from a company I used to respect more. It opened with Hi {{first_name}}, — the template variable literally not rendered. Their automation didn't even bother to substitute my name before hitting send.
So I built a thing that doesn't bother either.
How it works
The classifier reads every inbound email landing in my Gmail primary tab, sorts it into one of fourteen categories, and routes it three ways:
- Bulk-template outreach — cold recruiter blasts, lead-gen pitches, SEO and link-exchange schemes, "you've gotta hear about these pre-hype startups" intros — gets a templated reply auto-sent in my voice. No human reviews it on my end, because no human reviewed it on theirs. They mass-emailed me; my system auto-replies. Symmetric.
- Anything that might actually matter — recruiter pings from companies I'd genuinely consider, real partnership conversations, fan mail, threads where there's history — gets drafted for me to review and send by hand. The thing that would make me feel terrible if I missed is exactly the thing the system protects from auto-anything.
- Pure noise — Substack digests, GitHub notifications I subscribed to, receipt emails — gets a label and goes silent. No reply.
The full technical build — Trigger.dev cron, Gmail OAuth, Claude classify-and-draft, a per-category graduation ladder where categories earn their way to auto-send — is in the previous post. This post is the other side: what it actually does to the inbox.
The receipts
Three real auto-sends from this week. Names and email addresses redacted; everything else is exactly what went out. None of these had a human in the loop on either end.
A DevRel role pitch for an MLOps platform I'd never heard of.
Inbound: 23 lines of templated MLOps copy that didn't reference a single thing I do. Outbound, three minutes later: I'm an engineer not in DevRel, link to my recruiters page, footer.
A lead-gen pitch from someone who'd "totally stalked my page".
Inbound: claimed they'd "never seen anything like" my work but couldn't name a single thing I'd built. Outbound, three minutes later: my site isn't a lead-gen funnel and I'm not looking to turn it into one.
The "the market has shifted, here are three pre-hype startups" namedrop.
Inbound: a16z and Sequoia and YC and three "pre-hype" startups stacked into a single paragraph. Outbound, five minutes later: same polite no, same link to my recruiters page, same footer.
All three went out without me reading either side. The senders got a polite, on-voice reply within minutes. I didn't lose a single context switch.
The footer
Every auto-replied message ships with this line at the bottom:
Your email was read and replied to by my homebrewed AI inbox-screener, because we don't template or mass-email people we really care about, right?
Right on the edge of saying get fucked politely. It only attaches to the bulk-template buckets — cold recruiter, lead-gen, SEO, the pre-hype-startup namedrop. Real personalized email never sees it, because the jab only lands when the sender's own message earned it.
If you didn't template me, you'll hear from me directly. That's the contract.
Today I go guerilla
I built this because I was tired of opening my inbox to find that someone spent four seconds copying a template and zero seconds reading anything about me.
I'm going to keep tuning this for my own use cases. If something like this would help you too, ping me — I'm collecting signal on whether this should become a real product or just stay my personal weapon against asymmetric email warfare.
Tell me you'd use this →The subject line on that link routes the message to a folder I check. Tell me the shape of your inbox, the kind of outreach you'd want auto-handled, and whether you'd pay for it. The more shape I get, the better I'll know if this should leave my laptop.
Discussion
Giscus