Well, seven years later, OpenAI released ChatGPT, and the web hasn't been the same since. AI flooded the space we had spent decades building for each other, one post, one blog, and one comment at a time.

The numbers are clear. Stack Overflow was once the beating, chaotic heart of programmer culture, where strangers helped strangers debug their code at 2am. Its monthly questions have collapsed from 109,152 in November 2022, the month ChatGPT launched, to 1,146 in August 2026.

Social media users complain, louder every month, about feeds drowning in AI slop: generic, soulless images and captions built to be scrolled past, not actually read or felt. What's coming out of Reddit or Pinterest especially hurts, since they used to show exactly what slop can't fake: a real person's specific obsession rather than a generic guess at what everyone might like.

Meanwhile, Google's AI Overviews above search results have cut click-through rates to websites. Pew Research Center found that 8% of visits with an AI summary led to such a click, versus 15% without one.

This year, Cloudflare, which routes a massive share of web traffic, confirmed that bots now generate more web requests than humans do. As of today, the rate is 58.5% by bots against 41.5% by humans, though the bots' share is growing quickly: check yourself.

And this is only the beginning. AI agent activity online is expected to rise sharply in the coming years. Dead internet theory, the idea that the web would one day be mostly machines talking to machines, is looking less and less like a theory.

But humans are a nostalgic species, and capable of wonderful resistance. In July, designer Eric Lu created Ghost Font, an anti-AI font only humans can read. Static, it looks like noise, but in motion it reveals words no image-based AI model can currently catch: Hello Human. Without exact instructions on what to look for, ChatGPT can't read it. Neither can Claude or Grok.

This story lit a small fire in me that grew brighter days later when I realised Lu wasn't alone.

A Brazilian studio and a Danish foundry built ShieldFont, a free, open-source typeface with a different trick up its sleeve. On your screen, the text reads exactly as the author wrote it. But behind the scenes, the code swaps out about a quarter of the words for others. Grammatically, the sentence still holds, so a scraper has no reason to flag it. But it no longer says what the author meant.

Neither trick will save the internet on its own. A screenshot of the rendered page, for example, still gets you around ShieldFont. But 11 years after that winter in 2015, humans are finding small, stubborn ways to resist being pushed out of their own internet.

If you're reading this, it might not be too late.