Which AI products haven't made it and which one would you return from AI graveyard? [report]
Since AI became publicly available (2022), almost every product started containing an "AI" label. It is like a trend (I would say a spreading virus.)
But not everybody can be successful.
Around 42% of corporate AI initiatives are abandoned. (S&P Global report says)
Problems include funding, technical problems, competition, scalability issues, or weak demand.
TechCrunch shared in the article ~10 AI projects that didn't go well:
OpenAI: Discontinued or absorbed standalone products like Atlas, Operator, DALL-E and Sora.
Apple Siri: Delayed its promised AI upgrade repeatedly, leading to a $250 million settlement.
Character AI: Bankruptcy following legal challenges, regulatory pressure, and operational difficulties.
Relay: Shut down after larger platforms integrated similar AI automation features.
Ollie: Closed due to weak growth, insufficient funding, and difficulty differentiating itself.
Microsoft Recall: Faced major privacy and security concerns.
Notion Mail: Scheduled to shut down as users increasingly preferred other AI email agents.
Humane AI Pin: Failed to deliver reliable performance and shut down its hardware business.
Rabbit R1: Struggled with poor reviews and unreliable functionality but continues development.
Huxe, Yupp and Figgs AI: Shut down due to competition, weak product-market fit, or unsustainable operating costs.
We are still talking about a bursting AI bubble, but I think that AI will remain with us for a long time (the same was with the internet).
Anyway, if you could arrange a comeback from the above-mentioned products, which idea would deserve a second chance to succeed?
I liked Yupp – comparing results from more AI models side by side and Humane AI Pin – a wearable for AI communication.

Replies
yupp's side by side model comparison is the one i'd bring back too. do you think it died because people just stuck with their one favorite model?
@neerajnathany Oh yeah, I loved comparing the side by side
@neerajnathany I think yes, when you have many options side by side, it is becoming complicated.
There was one here called Tonkotsu by @derekattonkotsu I loved the idea and the philosophy.
I think the tech wasn't ready for the scope of tasks that it wanted to do. and it had this mix of Jupyter notebook meets notion but for docs. Which while my brain didn't exactly mesh with, it was and is a way many developers love to work.
I too write big detailed specs but in a slightly different way.
I think the team pivoted but the UX idea was there, the models hadn't caught up.
I'd love to see the philosophy reborn, especially since the AI bubble popping means tokenmaxxing is over. but these managers and orgs have become addicted to the velocity, the feel the need, the need for speed.
Even if Goose is gone they want a golden one.
SO we will need to engineer reliability out of LLMs or switch to world models and JEPA... and maybe maybe it won't need scaffolding, engineering, etc.
Model capability as drivable by humans has plateaued.
Raw model capability may have as well.
I think the focus is and should be on open, flash sized models.
GLM 5.3 flash was a revelation, and so was Deepseek 4.1 Flash 1017.
Liquid AI has been killing it with LFM3.
Qwen is the goat of 27B.
I think we will se a lot more development in small models.
@derekattonkotsu @build_with_aj Is the website already down for that? I would like to learn more about the thing you described.