I built an AI agent that automatically commented on HN. Here's what I learned
The system uses browser automation (Playwright) to navigate HN, read posts, decide which ones to engage with, draft comments in my writing style and submit them. It tracks what it's posted to avoid duplicates and sends me Slack notifications for each comment so I can check it.
I ran this for a few days. Some comments got decent upvotes (20+). Then someone called me out - they noticed my comments were posted exactly 45 seconds apart across different threads. Obviously a bot pattern. It would have been easy to build out a guardrail for this, create a new account, and make the agent run again. Instead, I'm stopping this experiment and would like to start a discussion with the HN community:
I'm genuinely curious how people feel about this. Here's my attempt to steelman both sides:
The case for allowing it: The comments provided genuine value. They were upvoted, sparked discussions, and answered questions helpfully. This is effectively generating new synthetic training data. Future models will learn from these interactions. If the output is indistinguishable from a thoughtful human comment, does the source matter?
The case against: It undermines trust. HN's value comes from authentic human discourse, not optimized engagement. If everyone did this, HN becomes a battlefield of competing AI agents, drowning out real humans. The comments may be "valuable" but they're also calculated - optimizing for karma, not genuine contribution. It degrades the social contract that makes online communities work.
As said, I won't be posting comments this way anymore. But to be honest, that doesn't really matter. If I was able to do this, anyone is. The genie is out of the bottle - this is trivially easy to build.
What can HN (and other forums) do to detect/prevent this?
I feel like we have a limited time window to figure this out. If it's ignored for long enough, AI generated comments _will_ become indistinguishable from human generated ones. I don't think there's a turning back after that.
No clue.
But if it becomes actually true that I can't distinguish between real and AI comments, then my personal solution is to stop bothering to read comments entirely.
Which, for HN anyway, would pretty much remove the entire benefit I get from it.
What's more concerning to me is despite all that and clear indications of it being a bot, dang seems to have ignored my comments and kept interacting with it two times:
https://news.ycombinator.com/item?id=46889769
If our own administrator of the website is getting duped, how are we going to be able to moderate effectively moving forward? Pointing it out multiple times ( y me and others) doesn't seem to have worked in this case.
If you notice an account that appears to be consistently pattern-matching to this, and have a minute to let us know, we'd appreciate a heads-up at [email protected]. We don't come close to seeing everything that gets posted here, but we do monitor the inbox closely (including fishing through the spam bin for real users) and we take these reports seriously.