
🍯 Honeypot v2: Crypto Boogaloo
Sixteen days of a fake crypto-finance startup: 95,231 SSH attempts from 4,641 addresses, a handful of guests who read the wallet seed page, and nobody who took the bait.

Sixteen days of a fake crypto-finance startup: 95,231 SSH attempts from 4,641 addresses, a handful of guests who read the wallet seed page, and nobody who took the bait.

Five of my seventeen tasks were the same job. I liked day lists before Claude — the dopamine does the work — so now it reads monday and hands me the queue.

I built a fake WordPress store, pointed it at the internet with no ads and no audience, and logged everything that came through the door: 51,561 web requests, 29,481 SSH connections, 805 addresses, and zero attackers who read the shop.

Most people use AI for memes and homework. The half that uses it to work faster produces the same work as everyone else. The panic is the loudest part, and the numbers do not back it.

Paying an AI per token is the lines-of-code mistake all over again — the real point is what the craft of engineering pays for.

The agent that rebuilt my server after a disk-killing outage restored everything I asked for — then put my media server on the public internet.

Every AI turn re-sends the whole conversation plus a system prompt you never wrote. That overhead is why the old prompt habits still save the most money.

Move fast and break things only worked because the person breaking understood what they broke. AI kept the speed and removed the understanding.

The 2026 small business ships an unreviewed AI app the way it used to ship admin/admin — and 45% of AI-generated code ships with a security flaw.

Seven AGI announcements in nine months, a Fireship video that almost got me, and the independent benchmark that did not.