Unpopular Opinion: Most 'AI Features' Are Worse Than The Old Button

I keep running into new AI features in apps and websites that make simple tasks slower, less accurate, and more frustrating than the old one-click button they replaced. What happened is that tools I used to rely on now add extra steps, guess wrong, or hide basic controls, and I need help figuring out whether others are seeing the same thing and how to work around it.

You’re right. A lot of these AI features replace a fast, deterministic action with a slow guess.

Old flow:
Click button.
Get expected result.

New flow:
Open AI panel.
Wait.
Read a paragraph.
Fix the wrong parts.
Try agian.

That is worse UX.

The core problem is fit. AI works best when the task is fuzzy, like summarizing notes or drafting text. It works poorly when the task is fixed, like sort, crop, export, search with exact filters, or filling a known form. If your old button had 98 percent success and took 1 second, replacing it with a 70 to 85 percent AI guess that takes 5 to 15 seconds is a downgrade.

You see this a lot in search. You ask for a setting or file. The app gives you a chat box instead of direct navigation. Now you need to describe the thing, then check if the answer is wrong. Same with photo apps. “AI enhance” often changes faces, text, or colors when all you wanted was one clean slider.

Best fix is simple. Keep both.

  1. Leave the old button in place.
  2. Put AI behind an optional tab.
  3. Measure task time and error rate, not hype.
  4. Let users turn it off.

If a feature adds steps, lowers accuracy, or hides controls, it failed. Fancy label or not.

Yep. A lot of it feels like product teams confusing “can use AI here” with “should replace the thing that already worked.”

I mostly agree with @viajeroceleste, but I’d push back on one part: sometimes the issue is not just bad fit, it’s bad incentives. AI features demo well. A one-click export button does not. In a meeting, “ask the app anything” sounds futuristic. “We kept the reliable menu” sounds boring. So users get the flashy version even when it’s clearly worse in daily use.

The most annoying part for me is loss of confidence. With the old button, you knew what would happen. With AI, even when it gets it right 8 out of 10 times, you still have to verify every single output. That means the mental load goes up. People underrate that. Speed is not just seconds, it’s how much brainpower a task costs.

Where AI does help is reducing blank-page friction. Drafting, rewriting, maybe rough organization. Fine. But if I’m adjusting settings, renaming files, cropping an image, or finding one specific option, I do not want “assistant behavior.” I want machine behavior. Exact, repeatable, boring.

Also, companies need to stop hiding normal controls behind the AI layer. That part is just hostile UX tbh. Add the smart tool if you want, but don’t make me ask a chatbot to do what used to be one click. Thats not innovation, thats removing a screwdriver and handing me a Swiss Army knife with a speaker in it.