What’s the most cringe-worthy thing you’ve seen in the AI agents space? My 3 top pics
Today, I’m doing a slightly more relaxed and bizarre corner.
The internet is full of things that are either amusing or scary, but mostly things that capture something outside the norm (and over time, even these weird things tend to become normalised).
What are the 3 strangest things in AI agents you’ve come across recently (or in general)?
Here’s my list:
An OpenClaw bot sent an inappropriate message to a boss on behalf of an employee via WhatsApp.
An AI bot paid for an overpriced course just to access the required information.
384 views
Replies
vibecoder.date
It was all over twitter, that exec who got her emails deleted by an agent.
https://x.com/FareaNFts/status/2026395423393591344 that's the full story.
Though if I put my tinfoil hat on for just a moment... negative PR stunt to allow META to catch up?
Massive attention on agents being unreliable, Presents the opportunity to introduce safe agents, since Meta sucks at models anyways.
Tinfoil hat off.
minimalist phone: creating folders
@build_with_aj I have a feeling that even people with some knowledge of tech cannot do anything about their uncontrollable bots :D we are so doomed :D
vibecoder.date
@busmark_w_nika At this point I might start ahving devices fully air gapped from any AI ussage, even local models lol
I sandbox everything I can but still, agent reliability is yet to be solved
minimalist phone: creating folders
@build_with_aj Did you see this? Probably switch from OpenAI to Claude.
At this point AI agents aren’t assistants… they’re unsupervised interns with WiFi.🤣 Waiting for the day my AI resigns on my behalf because it found a better founder to work for. 🤖
minimalist phone: creating folders
@shubham_saurave But they are cheap!!! interns :DD
Your #1 is actually the "good" failure mode — at least it was loud and visible. The cringe I genuinely dread is the opposite: the agent that silently completes the wrong task with total confidence. No WTF moment, just a polished email sent to the wrong 5,000 people, or a hotel booked on the wrong dates, discovered only after checkout. Loud failures get caught and fixed. Quiet competent ones don't surface until it's too late. Does the silent-but-wrong failure keep you up more than the embarrassing-but-obvious one?
minimalist phone: creating folders
@giammbo do you also have more cringe OpenClaw moments? :D
@busmark_w_nika Ha, oh yes — my favorite: asking it to 'quickly verify my API keys' and watching it helpfully test them by firing 12 real calls to a paid endpoint. Very thorough. For Aitinery the nightmare scenario is the agent that books the right hotel — just in the wrong month, with full confidence. Did the cringe moments end up shifting *how* you prompt, or mostly just what you give agents permission to touch?
@busmark_w_nika @giammbo Or did the cringe moments make you think about placing a human checkpoint?
@agraciag Yes — the cringe catalog is basically a checklist of where checkpoints should have been. The pattern I keep coming back to: a checkpoint only helps if it fires before the point of no return, not after the surprising outcome. The tricky design question is figuring out which action in a chain is actually irreversible — that's the one that needs the gate.