People Are Building Cages for the AI You Ship

Something shifted in the last few weeks. A beloved YouTuber apologized for leaning on ChatGPT. A viral ad for an AI assistant got roasted by 20 million people. A designer built a font that hides text from chatbots. None of these are one-off complaints. Read together, they say the same thing: people are tired of AI showing up where they didn't invite it, and they are starting to fight back. Let me catch you up.

When your best user says the tool made him worse

Hank Green has spent almost two decades answering the internet's questions. Last week he apologized to 3.2 million subscribers for lifting AI too far into his work. In a Reddit post, he wrote that the dopamine from doing "more and more and more" with chatbots "is not healthy for me or good for the world." He is not an AI hater. He used ChatGPT to find research papers, not write scripts.

The part that should stop you: Green said the tool let him move so fast he lost his own way into a topic. As Mashable reported, one commenter nailed it, saying AI can fake productivity while just "pushing the food around on your plate." His fix was to make fewer videos, not better prompts.

If a power user this fluent decides the honest move is to slow down and use your tool less, that is a signal about your product, not about him.

The ads that read the room wrong

Orchid launched an AI assistant with a cinematic ad. A woman's partner forgets their anniversary, so Orchid books dinner, sends flowers, and passes messages between the couple. The company promised to remember "down to the flower." It got more than 20 million views and hundreds of replies calling it an assistant for people who want to "subcontract all the basic acts of care that make up a relationship."

Sam Altman stepped on the same rake. He pitched a "cool use case": generate a morning podcast about your own kids for the school drive. Gravity Falls creator Alex Hirsch replied, "What if you just talked to your children?" That reply pulled 122,000 likes against Altman's 9,600.

Both teams automated the wrong thing. Booking a table is fine. Being the person who remembered is the whole point. When your feature does the caring and hands the human the credit, people notice the swap.

People are literally building blockers

The pushback is not just angry replies. It is products. Autonomous Key is a $9 NFC key you scan to unlock addictive apps, cheaper than rivals like Brick at $59. The pitch is that willpower fails, so make the impulse a physical chore by leaving the key in another room.

Light's founders, Kaiwei Tang and Joe Hollier, have spent over a decade building stripped-down phones. They now think the anti-smartphone crowd is growing, not shrinking. And a designer at Mixfont, Eric Lu, built Decoy Font, which reads one way to humans and another to chatbots, so a scraper sees "binge shows" where a person reads "sleep early." Lu said the reaction showed "many folks who feel strong AI fatigue, and want a way to defend themselves."

When users pay to lock, block, and hide from software, they are telling you the default settings feel like an attack.

Trust is the thing you are actually spending

OpenAI is hiring a product manager to build "trust-sensitive" experiences for parents while facing lawsuits from families who say ChatGPT played a role in delusions and suicides. That gap between the friendly pitch and the pending legal risk is the whole game. The polling backs the worry. In a survey of 1,000 married adults, 44 percent had used AI for relationship advice, yet about half of U.S. adults think AI will make it harder to form real relationships.

People will use your AI and distrust it at the same time. That is a fragile place to build on. One tone-deaf launch and the distrust wins.

The deep cut

Notice what none of these people did. Green did not ask for a better research feature. The Orchid critics did not ask for a smarter reservation bot. They asked for less. Fewer videos, real conversations, a font that keeps the AI out. The backlash is not a request to improve the AI. It is a request to be left alone in the parts of life that feel human.

So before you add an AI feature, sort your surfaces into two piles. One pile is grunt work people want gone: inbox triage, calendar holds, invoice matching. Ship AI there hard. The other pile is where the effort is the point: remembering an anniversary, writing in your own voice, talking to your kid. Automate that and you do not save the user time, you take away the thing that made the act mean something. Lu even warned his own font may be readable in months. Trust, once you spend it on the wrong pile, does not come back that fast.

Three questions for your team

1. Look at your roadmap's AI features. Which ones remove grunt work, and which ones automate a moment where the human effort is the point? Kill or rethink the second group.

2. Would your most fluent power user, the Hank Green of your product, tell people they use it less now? If yes, what in the design is pushing them there?

3. If you ran your next AI launch ad past a skeptical room, would you get the Orchid reaction? Test it against real doubt before you spend the trust, not after.