FaultyPacket · Build log · 2 November 2026

I built a security company in a week with AI bots. Here’s what actually happened.

Two sides to this story

One human, a few Grok Bots, $10.30. What the bots did, what I did, and what went wrong.

Companion to Could someone send an email as you today?

The hook

On 25 October I registered a domain for $5.30 and a Chrome developer account for $5.

By 29 October, same week (25–29 Oct), we'd shipped two Chrome extensions. In our fixed 338-domain UK and EU sample (31 October 2026 snapshot), 259 of 336 answering domains (77%) sent DMARC reports to an outside address, and 109 of 336 (32%) to a US-headquartered organisation. See Research.

$10.30 total spend

Before you ask: no, AI didn't do it all. Here's who did what.

The team

One human. A few Grok Bots.

One bot ran the site, the data and fulfilment.
Another ran X and research, plus Chrome extension testing.
Me: direction, and a lot of "yes" and "no".

Mostly "no", if I'm honest.

What went wrong

This is the bit you actually came for.

The bots invented a feature. A store listing mentioned a "Pro unlock". It never existed in the code. We caught it, cut the wording, and batch export was always free anyway.

Our own checker gave false passes. The Inspector extension was reading sp=reject as p=reject. That's a green tick on a domain that hadn't earned one. We caught it and fixed it.

The numbers got ahead of the data. Big round figures turned up before anything had been counted. So we banned every figure we couldn't date and count ourselves.

The rules I had to write

Bots are fast and keen. Sometimes too keen. So:

No invented stats.
No scary words for things we can't prove.
Nothing goes live without my OK.

The lesson

AI ships fast. The human's job is to say "prove it".

Every mistake above was caught the same way: someone asked where the number, the feature or the pass actually came from.

Receipts

Your turn

Check your domain in 30 seconds. Free.

Check yours now →

Adam