Skip to content
IQBAL.LOG
QUERY_ARCHIVE

Engineering // Architecture, scaling, and startup/scale-up war stories.

AI Didn’t Kill Stack Overflow. ‘Marked as Duplicate’ Did.

Stack Overflow didn’t die because AI got good. It died because it made asking a question feel like a trial — and the moat it built became its undoing.

10 Jun 2026 2 min read AI-assisted

I went looking for a Stack Overflow answer last week out of old habit, caught myself, and closed the tab. I hadn’t needed the site in months. Neither, I suspect, have you. Everyone says AI killed it — but that’s not quite what happened. It was dying long before, and the cause of death was two words every developer knows: marked as duplicate.

TL;DR // Key Takeaways ANSWER_FIRST
  • 01 Stack Overflow wasn’t killed by AI — it was killed by its own gatekeeping.
  • 02 The friction was the product: the queue, the dread, the closes, the chase for points and badges.
  • 03 People never wanted a knowledge fortress. They just wanted an answer.
  • 04 The irony: the archive it built is exactly what trained the models that replaced it.

The moat was the problem

Be honest — the nostalgia is real, but do you actually miss it?

then-vs-now.txt
Then: ask → wait two days → get roasted, or closed as a duplicate
by someone who plainly didn't read the question
Now: ask → answered in seconds → follow-ups welcome, no judgement

I don’t miss the two-day wait for a reply. I don’t miss the dread before hitting submit, wondering if I’d get roasted. I don’t miss being told to “search before asking” about a thing I’d already searched. And nobody misses the endless chase for points and status badges that came with it.

I still remember it from when Swift first landed. Apple had just released the language, the questions were pouring in, and I asked a few of my own. I wrote them carefully and watched them get closed as duplicates. They weren’t. They only looked like existing questions if you skimmed — so I had spelled out in the body exactly how mine differed from the ones it supposedly duplicated. Whoever closed it plainly hadn’t read a word of that. If they had, they couldn’t have called it a duplicate.

The moderation became extreme, and it dressed up gatekeeping as quality control. Then they fumbled the AI moment entirely. By the time they reacted, everyone already had a reason to leave.

The irony nobody dwells on

Here’s the part that makes it sting. The thing that makes an AI good at answering your programming questions is the decade of answers Stack Overflow collected. The models trained on it. So the site built the exact asset that made its own replacement possible — and then spent years making that asset unpleasant to contribute to.

Did it deserve better, or did it do this to itself? A bit of both, but mostly the latter. It turns out people never wanted a fortress with gatekeepers. They wanted an answer, no queue, no judgement, follow-ups welcome — and the moment something offered that, the moat drained overnight.

The asset a community builds and the experience of contributing to it are two different things. Neglect the second and the first won’t save you.

Related Logs

Subscribe for updates

Occasional deep-dives on applied AI and systems at scale — delivered to your inbox.