Voice Interfaces Need Absolute Reliability

The Apple Watch failed (for me). It relied heavily on voice interaction through Siri, and Siri is both underpowered and flaky. That combination is fatal for an interface that can't fall back to typing.

I forgive Claude for being flaky because it's overpowered. When you fail to execute something, I know you could probably do it if I rephrased or tried again, and the payoff when you work is worth the friction. Siri is the opposite—underpowered and unreliable. You can't trust it to handle even simple commands consistently, and when it does work, the capabilities aren't impressive enough to justify the mental overhead of wondering whether it'll work this time.

For voice interfaces to succeed, reliability has to come before capability. People will accept a limited feature set if they can trust it completely. My G Shock doesn't do much, but I never question whether it'll work. That trust is what makes limited functionality acceptable.

Voice interfaces break trust in a way visual interfaces don't. When you tap a button and nothing happens, you tap again. When you speak a command and nothing happens, you feel stupid. The social awkwardness of talking to a device that doesn't respond makes unreliability exponentially more frustrating than the same unreliability in a touch interface.

The Apple Watch needed Siri to be bulletproof. Not smarter, not more capable—just completely trustworthy. If every voice command worked every time, even a limited set of commands would have been compelling. Instead, Siri's unreliability meant people stopped trying to use voice entirely, which eliminated the one interaction mode that made sense on a wrist-worn device.

Ambient computing depends on trust. You can't think about whether an interface will work when that interface is supposed to fade into the background. Voice has to be so reliable that you stop thinking about it. Until that happens, it's just another flaky feature that drives people back to simpler, more predictable tools.