Harshita Nimbal explains: Designing the Right Input for AI (Voice, Text, or Neither)

The same AI feature can talk to you, wait for you to type, or say nothing at all. Harshita Nimbal, a product designer, asks what decides which of these is right. Her answer is that the best input is not voice or text but whichever fits the moment, and the video builds that case with real products.

Three modes, three jobs

The video frames the choice simply. When your hands or eyes are busy, talk to the AI. When you need exact words or privacy, type to it. And sometimes the right choice is to say nothing and let it act. The rest of the video gives an example of each.

Voice when hands and eyes are busy

Google Assistant's driving mode starts with "Hey Google, let's drive" and hands every task, including texts, calls and music, to voice alone, so hands never leave the wheel. The video points out that this is not a stylistic choice. It is the only input that works when eyes and hands are already occupied with driving.

Text when wording must be exact or private

The opposite context is an open office where someone is writing code. GitHub Copilot suggests code inline as dimmed text. Nothing is spoken aloud, and nothing runs or saves until you press tab. Typing keeps exact syntax exact, and it keeps your prompts private from the desk next to you. Voice would be slower and more awkward here.

Neither: invisible automation

The third mode is easy to miss because it is invisible by design. The AI just acts and nobody has to say a word. Gmail's spam filter blocks more than 99.9 percent of spam, phishing and malware, across billions of emails a day, without you typing a search or giving a command. You notice it only on the rare day it is wrong. The Nest learning thermostat watches what temperature you choose, and when, for about a week, then builds its own schedule, with no app to configure and no voice command.

Why matching matters

The wrong input adds a tax every single time. Fumbling to type while driving is dangerous, speaking a private query aloud is embarrassing, and asking permission for something trivial is annoying. Voice in a quiet office is disruptive, while voice with busy hands is right. Typing while driving is dangerous, while typing when precision or privacy matters is right. Acting without asking on something big is unsettling, but on something small and reversible it is exactly right.

The "neither" mode has a real cost, because it usually means always listening. The video cites Apple's agreement to pay $95 million to settle a lawsuit after Siri was found to record private conversations by accident, including medical details and business calls.

Three decision questions

Before defaulting to a chat box or voice assistant, the video says to ask three questions. Are the user's hands or eyes busy? Does the wording need to be exact or private? Is the action small and reversible enough to skip asking entirely? The answers pick the input.

Key takeaways

• The best AI input is whichever fits the moment, not a fixed default.
• Voice suits situations where hands or eyes are busy, such as driving.
• Text suits cases where wording must be exact or private.
• "Neither" suits small, reversible actions that can happen invisibly.
• The wrong input imposes a cost every time it is used.
• Always-listening designs carry real privacy risk.
• Three questions about busy hands, exactness or privacy, and reversibility choose the mode.

Try it yourself

Pick one AI feature you use and ask Claude whether your hands or eyes are usually busy when you use it, whether the wording needs to be exact or private, and whether the action is small and reversible enough to need no input at all. Then ask whether it is using the right mode today or just the one its designers defaulted to.