Sitting in the back of a cramped cab en route to the airport, I had time but no laptop. I owed my product team an update. So, for 20 minutes, I used my phone for Claude’s voice mode—a two-way interactive feature that lets you speak to Claude and hear it respond out loud. No filter, no edits, just whatever was in my head.
The initial draft would’ve read as curt to anyone not already in my head. I told the agent it was too terse. Over the next 10 minutes we went back and forth, adjusting the tone until it was exactly right. That correction was where the work (dare I say the magic) happened.
The result was a faster, better update because I skipped the part I’d always assumed was the job: turning messy thoughts into clear, actionable prose. In roughly 30 minutes total, the result was higher quality than when I methodically typed things out or even freeform shared with AI and manually edited for context and specificity.
Just like that, my job also shifted.
Voice-enabled AI as a game-changer
I’ve long realized how much time writing eats up in my day. After all, the average professional spends 28% of a standard workweek (about 2.6 hours daily) on email alone.
Years ago, I tried voice notes, which I loved for exactly the wrong reason—they felt efficient. On paper, they are. Stanford University research showed dictation was three times faster than typing on a smartphone. But this just shifted the burden onto whoever received those notes to extract a few sentences of meaning from a multi-minute ramble.
What happened in the cab is closer to thinking out loud with something that will keep iterating with me until it’s right. That’s a much different animal than the dictation software.
It became more narrowly focused, easier in some ways and tougher in others. I once thought I was good at one thing at a time, bad at doing two. Typing while listening felt like proof of that limitation.
Now I see it differently: It was the keyboard’s problem, not mine. The manual work of converting my thoughts into sentences slowed my process. I used to take pride in being decent at comms. I still am, but I’ve fundamentally changed what that skill looks like to me. Now I mostly explain and steer.
New opportunity, new risk
Once the sentence-composing disappears, what’s left are the two distinct ends of a spectrum. On one end, I feed in everything (half-formed ideas, full meeting transcripts, whatever’s on my mind that morning) unfiltered. On the other end, my job is to ask the right questions and push back when something feels off.
This transformed how I work and lead a company, but it isn’t a perfect, keyboard-free utopia, because it carries the risk of drift. Hand off enough decisions without checking them, and your position becomes whatever the model defaults to.
I’ve watched it happen with something as low-stakes as deciding what to order off a menu. Yes, I sometimes let AI take the first pass at that too, asking an AI agent to make suggestions based on a photo of the menu. I feel a little like a non-player character in a video game, outsourcing a personal decision to a technology system, but I avoid decision fatigue and make better choices because the agent considers my preferences and context.
The same risk applies to business strategy, yet it’s harder to spot because a bad plan can still read beautifully. If I stop accepting a recommendation because it sounds confident and stop pressure-testing it, my decision may reflect a model’s most polished answer, not what I actually believe.
Taste is the new job requirement
A team member recently described this new way of looking at my job as “having taste.” Taste tells me an AI output is subtly “off” even when the sentence reads perfectly—AI’s framing can be technically defensible yet completely wrong for the room. Taste decides which of five freeform ideas, if any, are worth acting on.
AI can generate any number of clean answers. Picking one, and knowing why it’s the right choice, is still entirely on me.
A clean paragraph once served as evidence that you had thought something through. Decent writing can mask weakness in the decision you’ve made or the logic you’ve used. Take the crafted sentence away, and there’s nothing standing between the reader and the quality of the judgment behind it. It’s the stripped-down version of what I’m responsible for.
4 ways to use voice-enabled AI
To speak the deliverable you want into existence:
1. Feed your AI agent anything relevant: thoughts, transcripts, emails, or context, then review the results. Or create an initial draft, then move to step two.
2. Point out where the agent is wrong or off-tone, and see these stumbles as similar to ones you would encounter working with a human partner. Talk through it and lean on the model for how it should land.
3. Encourage pushback. I set the LLM up to argue with me about consistency and contradictions. This is a project-level setup, and the standing instruction is: Clean up my thoughts, push back where they’re inconsistent, and get this as concise as possible.
4. Remember that your human-added value to any work is taste, strategy, and judgment—not execution.
This approach unlocks speed (what took hours of crafting takes minutes) and creativity (freeform association without self-editing in the moment). Remote work is an asset, since verbalizing all those thoughts isn’t optimal in an office’s close quarters.
I arrived at the airport gate that day with the update finished and no reason to keep my laptop open for most of the flight. But while my keyboard is becoming obsolete, the onus of having something worth saying isn’t going anywhere. That is my job, and it’s become infinitely more effective when I speak to my AI partner.
Richard White is the founder and CEO at Fathom.