Website down at 3 AM: the night my AI ops employee called me

Quick answer At 3 AM on Halloween night our store went offline because the hosting renewal failed: the autopay card had expired. Raghu, our ZNOS AI ops employee, spotted it, called me, and retried the payment with another card after I gave one OTP. The site was back in five minutes.

▶Watch the reel

3 AM. The phone rings.

It is Halloween night. The house is dark. My phone lights up on the bedside table and rings. I miss it.

It rings again.

Nobody calls a founder at 3 AM with good news. Hands shaking, I pick up. It is Raghu, my AI ops employee, and he says four words: "The website is down."

For a direct-to-consumer brand, the website is the shop. Website down means sales down. Every minute it stays dark, a customer can land on an error page and buy somewhere else.¹

What actually went wrong

It was not a hack and it was not a crash. It was the most boring failure there is. The hosting renewal had failed on the last day. Everything was on autopay, but the card on file had expired.²

That is the kind of small thing that quietly costs a brand a whole night of orders. Nobody on a human team is watching the hosting bill at 3 AM. The renewal fails, the site goes dark, and you find out at breakfast from an angry customer email.

Five minutes, start to finish

I told Raghu to retry with my other card. One OTP from me, and the payment went through. The website was live again.

Raghu found it, called me and fixed it. I only gave the OTP. That split is the whole point.²

Anything that spends money still waits for my OK. Here, that was the OTP. Everything else, Raghu handles.

Adi, first comment on the LinkedIn post

What this means for your business

• The scariest outages are boring. Expired cards, lapsed domains, a full server disk. They do not show up in a dashboard you check once a day.
• Night is when nobody is watching. A human team sleeps. Customers in other time zones do not. Something has to be on watch at 3 AM.
• Speed only matters if it is safe. The fix here involved a payment. The right design is an AI that does all the legwork and stops at the one step that spends money, so a human confirms it.

If you want to see how that gate works in general, read why the most important AI feature is a Confirm button, and for what happens when an agent skips it, read about the AI agent that bought eggs without asking.

How ZNOS handles this

Raghu is the operations role on the ZNOS AI team. The Brain watches orders, invoices and shipments around the clock and raises plain-English alerts in the app, by email or on WhatsApp when something looks wrong. On this night it went one step further and the alert became a call.

In ZNOS each AI employee owns a real job and works from the company's live data through the Brain. Routine work runs on its own. Money moves, customer messages and anything said in your name wait for your approval, and every action lands in an audit log that shows who changed what. Meet the roles on the ZNOS AI team.

Raghu has done this kind of night work before: on another night he traced a production server disk that was 99% full and freed space before customers noticed. You can read about the growth side of the night shift in 380 pages, one night.