You know that sinking feeling when you find a feature you thought was gone, but it was hidden all along? That moment of oh, it was there the whole time—followed by a quiet frustration that no one told you.
I just had that moment with Codex. And then I saw Doubao quietly fix the exact opposite problem. Two small changes. One massive lesson about how AI products are failing users.
The best features are the ones you never knew you needed — until you accidentally find them. But if you find them by accident, the product is already broken.
Here’s what happened. Codex used to have a prominent ‘New Chat’ button. Then it disappeared. I assumed the feature was killed. Then, months later, I hovered over a tiny icon and found it tucked inside a submenu. The feature was still there, better than ever—but I had been living without it because the design team assumed I’d notice a small icon.
Meanwhile, Doubao’s recording feature had the opposite problem. The old version slammed a huge floating window over everything you were doing. It was information-rich but context-blind. I stopped using it. Then they shrunk it to a tiny status bar—less intrusive, more useful. If your product changes every week, your users need a treasure map, not a one-time tour.
Most product teams treat feature discovery as a one-time onboarding problem. Show the user once, and they’ll remember forever. That’s a lie. AI products evolve fast. Updates happen weekly. What got buried in a release note last month is now a crucial workflow. And users don’t re-read release notes.
Here’s the paradox: A feature that’s hidden for cleanliness is effectively deleted. A status that’s visible for awareness is effectively disruptive. The same design must be both prominent and invisible, depending on context. That’s not a template problem. That’s a strategy problem.
The solution isn’t adding more onboarding tours. It’s building re-discovery into every update. When a feature moves, highlight it. When a status changes, acknowledge it. Don’t assume users will find it—they won’t. They have other things to do.
I saw this firsthand with Codex: I used the feature daily before it moved. After, I abandoned it for weeks. That’s a direct loss of utility. And for Doubao: I literally wrote a blog post complaining about the popup. They fixed it. Now I’m back. That’s a win—but it took a loud complaint to get there. How many users just silently left?
If you’re designing AI products, stop optimizing for the first time someone uses your product. Optimize for the hundredth time. That’s where the real value lives, and that’s where most products fail.
So here’s the takeaway: The next time you update your product, ask yourself—how will someone who’s been using this for six months know about this change? If you don’t have a good answer, you’ve just made a $1 billion mistake.
FAQ
Q: Isn't it better to hide features for a clean interface?
A: Clean interfaces are great—until they make features invisible. The trade-off is between aesthetics and discoverability. The best approach is progressive disclosure: show the feature when it's relevant, hide it when it's not. But never hide it without a path back.
Q: What practical steps should product teams take?
A: For every update, create a 're-discovery moment': a subtle highlight, a tooltip on first hover, or a one-time notification for existing users. Also, track feature usage before and after changes. If usage drops, you've hidden it too well.
Q: Isn't it possible users don't want to discover new features at all?
A: Some users don't. But that's a design failure, not a user preference. If a feature adds value, it's your job to make it findable without being annoying. The solution is contextual cues, not constant popups. Let users opt in to exploration.