Typing is the new bottleneck: why voice-first AI wins

Typing used to be the work. Now it's the toll booth.

For thirty years, typing was the entire interface to a computer. You typed code, you typed words, you typed commands, you typed search queries. The work and the typing were the same thing.

AI broke that bundle. The work is now the thing the AI does for you. Typing is just the toll you pay to get there. And like every toll, it adds up the more times you cross.

The hidden cost of one prompt

A single prompt feels free. You're already at your keyboard, you have a thought, you tap it out, you read the reply. Easy. The cost is invisible because you only feel it once.

Now zoom out to a workday. A product manager prompts the AI to summarize a Slack thread, then to draft a doc, then to compare two pricing approaches, then, in a meeting, to remind everyone what the team decided last week. Each prompt is small. The total is half the day spent in the prompt loop instead of the work.

The cost of typing is not the typing. It's the four other things that happen around it:

1. Context switch out: you stop the real work to phrase the request.
2. Cold start: you re-explain context the AI should already know.
3. Wait: you watch a cursor blink for a few seconds at a time, dozens of times a day.
4. Context switch back: you re-enter the work, often having lost the thought you started with.

Microsoft's Work Trend Index has tracked this for years: the average knowledge worker is interrupted every few minutes and rarely gets a full hour of focus. The prompt loop is the newest interruption, and we built it ourselves.

Why typing breaks down hardest in meetings

Solo work survives the prompt loop. You're already alone, the only person you're stalling is yourself, and a few seconds of latency is fine. Meetings are different.

In a meeting, every second of typing is a second the room stops moving. The strategy session pauses while someone formats the prompt. The client call drifts off-topic while a teammate scrolls back to find the right doc to paste in. The design review goes quiet while two people, on different keyboards, prompt the same AI in slightly different ways.

You can feel this in your body when it happens. The room sags. Someone fills the silence with a joke. By the time the AI replies, the question has changed.

The fastest meetings I've sat in this year had no typing in them. Someone said the thing, the AI did the thing, the team kept talking. The slowest meetings had four open laptops and a shared frustration that nobody could quite name.

That frustration has a name now. It's the typing bottleneck.

What voice-first AI actually changes

Voice-first AI is not "AI plus a microphone." It is a different shape of tool, with different defaults.

Listening replaces prompting

The AI hears the conversation as it happens. There is no input box because the conversation is the input. Whoever says "we should pull the Q3 retention chart" is, by saying it, also asking for the chart.

Action replaces reply

The output of a prompt is text in a chat window. The output of voice-first AI is the actual work: a chart on the screen, a doc shared to Notion, a follow-up filed in Linear. The reply, if there is one, is incidental.

Context replaces re-explanation

You don't tell a teammate, "as you may recall, three weeks ago we discussed pricing." You just say "the pricing thing." Voice-first AI works the same way. It already heard the earlier meeting. It already knows what "the pricing thing" is.

Silence replaces small talk

Good voice-first AI is mostly quiet. It speaks when the room needs it and shuts up when it doesn't. That sounds simple, but it's the thing chatbots fundamentally cannot do, because the prompt loop demands a response to every message.

The teams that already moved past typing

You can see the shift if you look at how the most AI-fluent teams run their meetings in 2026:

• Sales: the AE focuses on the prospect's eyes, not the keyboard. Pricing, feature comparisons, and customer references show up on screen as they're discussed, not after.
• Engineering: the staff engineer doesn't open a separate chat window during incident reviews. Postmortems get drafted in real time and the on-call rotation gets updated before the meeting ends.
• Design: review sessions stay in the file. References, comps, and prior decisions surface as the conversation moves, not in a Slack message at 6 p.m.
• Founders: investor calls become conversations again. Pipeline numbers and product clips get pulled while the founder is still answering the question.

None of those teams are typing less because they're slower. They're typing less because the keyboard is no longer in the critical path.

Where typing still wins (and should)

Voice-first does not mean voice-only. There are tasks where typing is genuinely the best tool, and trying to voice-ify them is a mistake.

Long-form writing is a typing job. The slow, deliberate pace of typing is the point. Same with code: the keyboard is a precision instrument, and dictating a function out loud is a worse experience for almost everyone.

Careful editing, table work, and any task where you need to point at a specific character on a specific line are all typing jobs. So is private journaling, where the silence of the keyboard is part of the value.

The right way to think about it: voice-first AI takes typing off the critical path of collaboration. It does not push typing off your laptop entirely.

How to tell if your team has a typing bottleneck

Three signals. If you nod at one of them, you have it. If you nod at all three, you have it badly.

1. The note-taker bottleneck. Your meetings depend on one person being good at typing, summarizing, and feeding the AI prompts. When that person is on vacation, the team feels it.
2. The prompt-writer cult. Someone on the team is unusually good at writing prompts, and slowly everyone else has stopped trying. The AI's quality is now a function of one person's keyboard.
3. The recap night. Meetings end, and then a second shift begins: typing recaps, formatting decisions, and re-prompting the AI to clean up its own output. This is the typing tax in its purest form.

If any of these are true, the fix isn't a better prompt library. It's removing the keyboard from the meeting.

What to look for in voice-first AI

The category is loud and a lot of products call themselves voice-first when they really aren't. A few honest tests:

• Does it act, or does it talk? Real voice-first AI files things, opens things, drafts things. If it only narrates, it's a chatbot with a microphone.
• Does it remember across meetings? The whole point is to skip re-explanation. If you have to brief it every time, it's a transcription tool.
• Does it know when to stay quiet? Good voice-first AI rarely speaks in a meeting. If it interrupts, it's optimizing for itself, not the room.
• Does the work show up in your tools, not its app? Voice-first AI should make your existing tools (Slack, Notion, Linear, Gmail) better. If everything happens in its own dashboard, it's just another silo.

Where relly fits

relly is built around exactly this thesis. The keyboard belongs in the places it's actually the best tool. Everywhere else, the work itself should be the input. relly listens to your meetings, holds context across them, and ships the follow-up work to the tools your team already lives in.

If your team is paying the typing tax, you can feel it. Early access is open through May 18, 2026, with 50% off for your first year. No card needed until launch.