Nobody Read the Robot's Vacation: My Month of Markov-Chain Auto-Replies

There's a specific kind of email you've written a hundred times and never once read. You know the one. "I am currently out of the office with limited access to email." It's the beige wallpaper of professional communication. It exists to be ignored, which got me wondering: ignored how much, exactly?

So I ran the experiment my site was basically built for. For one month, every auto-reply from my work email was written by a Markov chain I trained on ten years of spam and discarded contact form submissions — the digital sediment that's been piling up in my various inboxes since roughly 2014.

The Training Data Was Beautiful Garbage

If you've ever dumped out a contact form's spam folder, you know the material writes itself. There was a man who'd been offering me "premium backlinks at friendship prices" since the Obama administration. There was a long, oddly poetic run of messages about solar panel installations that always began mid-sentence, as if I'd walked in on the conversation. There were at least forty variations of "Dear Webmaster, I loved your blog post" — a compliment so generic it loops back around to being a threat.

I fed all of it, plus my own old auto-replies, into a simple Markov chain. No fancy language model. I wanted the specific dreamy incoherence of spam, not fluency. The output looked like this:

"Hello! I am currently out of the office with limited access to premium backlinks at friendship prices. For immediate assistance, please walk in on the conversation or reply DEAR WEBMASTER to unsubscribe from this vacation. I will return Tuesday, or upon receipt of your solar panel inquiry, whichever comes first."

Honestly? Reads better than half the out-of-offices I've received from actual executives.

The Rules of the Game

The parameters were simple. The bot handled every auto-reply for four weeks. No editing, no softening. If the chain produced something alarming, it went out anyway — that's the tax of doing science on your own career. I flagged myself to check for anything legally dangerous (accidentally promising someone a refund, threatening a vendor with "friendship prices") and gave myself a one-word veto: only if the email could get HR involved.

The veto fired twice in a month. Once for a message that promised a client "unlimited access to my body of work, forever," which, fine.

Nobody Noticed. I Mean Nobody.

Here's the part that genuinely rattled me: thirty-one days, roughly two hundred auto-replies sent, and the only responses I got were three out-of-office auto-replies back. Machines talking to machines, politely acknowledging each other's fabrications. One of them was a vacation responder that said "I'll be sure to follow up when I return." It will not follow up. It was never going to follow up. We both knew.

My coworkers found out after the experiment ended, when I showed them the logs during lunch. The reactions split perfectly down a line I did not expect:

• The engineers thought it was hilarious and immediately asked if they could train one on their spam.
• The account people were horrified, then quiet, then one of them said the thing I've been chewing on ever since: "Wait, so for a month, nobody needed me?"

She meant it as a joke. It landed like a eulogy.

The One Human Who Read It

One reply came from outside the company — a vendor I'd exchanged maybe six emails with over two years. His full message: "Just want to flag that your auto-reply has gotten weird. Hope you're okay."

He's the only person in a month who read the wallpaper. I bought him a coffee when our contracts overlapped at a conference in Denver. He asked what a Markov chain was. I explained. He said, "So it's basically a parrot with a head injury," which is the best definition I've heard and I'm stealing it permanently.

What the Static Taught Me

We talk a lot about how nobody reads the terms of service, but we've built a whole professional culture where nobody reads each other either. The out-of-office email is a ritual performance — we send the incense smoke of "limited access" and "prioritizing urgent matters" and everyone nods at the smoke. The content is decorative. The existence of the reply is the whole message. My spam bot understood this instinctively. It produced perfect ritual objects: warm, structured, and completely empty.

There's something almost comforting in that. The machinery of corporate communication runs on vibes so thick that a head-injured parrot can operate it undetected for a month. It also means every carefully worded auto-reply you've ever agonized over — the tone, the punctuation, whether "Cheers" is too casual — went into the same void mine did.

The bot is retired now. It lives in a folder on my site next to the loading bar that never finishes and the guestbook nobody signs. But I'll be honest: I kept one of its emails in my drafts. "I will return Tuesday, or upon receipt of your solar panel inquiry, whichever comes first."

Some days that's the most honest thing in my inbox.