TechCrunch ran a piece this morning on Wispr Flow’s India push. Their CEO Tanay Kothari keeps repeating the phrase “whisper-filled office of the future,” and the pitch is straightforward: voice AI eats the keyboard, knowledge work gets faster, India becomes a giant beta.

I read it twice. Voice has crossed some quiet line over the past year. Wispr Flow, Aqua, even Apple’s built-in dictation on Sequoia are now good enough that I let them write my first drafts most mornings. But my workday is not faster, and after a year of watching this play out I finally have a clearer sense of why.

The keyboard was never the slow part

Watch anyone work in a browser for an hour. They type, sure. But most of their time bleeds out into other things: clicking into the right sub-tab of Salesforce, scrolling a Notion page to find the row that needs updating, tabbing back to Gmail to copy a customer address into a Stripe invoice form with its own three-step shipping flow that nobody enjoys.

Typing is maybe 15% of that hour. The rest is navigation, context-switching, and the small humiliation of remembering where the hell the text needs to land. So when Wispr replaces typing, it speeds up the 15% and waves at the 85% from across the parking lot.

Forty seconds of plumbing

You dictate a reply to a customer. Wispr turns your voice into text in a floating panel. You copy that text. Open Gmail. Find the email. Hit reply. Paste. Fix the paragraph break the dictation tool stuck in the wrong place. Send.

Voice took eight seconds. The plumbing took forty.

That gap is the productivity illusion the voice-first companies are selling, and I think it’s a big reason the whole category never quite breaks out the way the demos promise. Because the demo always shows the voice. And the demo never shows what happens to the text afterwards.

The tab is the actual interface

Our bet at dassi has been the opposite of Wispr’s: the bottleneck is not the input mode, it’s the gulf between user intent and the authenticated SaaS surface where that intent has to actually land, and voice tools mostly cannot cross that gulf because they live outside the browser.

Your browser already holds cookies for forty SaaS tools. It knows which Salesforce account you have open, which Linear ticket you’re reading, which row of the Airtable you just clicked, and roughly how your team’s reply style has drifted across six months of sent mail.

A browser agent that runs inside that tab can take an instruction like “reply to this customer and tell them we’ll ship Monday” and actually click reply, write the draft, scroll up to verify the shipping address from a thread two weeks back, and hand you something you can ship with one keystroke. Because it sees what you see. Because it is logged in as you. Because it does not have to copy text across the app boundary that was the slowest part of the whole job to begin with.

So voice can be part of the input layer, fine. But the execution layer has to live where the work actually happens, and that place is your already-signed-in browser tab, not a floating panel that hands you a paragraph and abandons you in the SaaS UI maze. I wrote more about this in the post on cloud browser agents and tab context.

Where voice still wins

Walking the dog. Driving. Cooking with greasy hands. Anywhere the keyboard is unavailable, voice is great. The category isn’t dead. It just solved the easier half and called it a day.

The plumbing is the product

Wispr Flow has built a beautiful microphone. Nobody in the voice category has built the part where dictated text knows which form, which field, and which signed-in app to land in. That part requires sitting inside the browser, reading the DOM, and acting on behalf of someone who is already signed in.

And that’s what dassi does. It lives in the side panel of Chrome, sees the page you see, and uses your existing logins to push the work through whatever SaaS UI you’re currently stuck inside. Grab it from the Chrome Web Store. Bring your own LLM key or sign in with an existing ChatGPT subscription.

A whisper-filled office sounds romantic. An office where the tabs do their own clicking sounds like Tuesday.