Talk, talk, talk: The rise of AI dictation tools at work
For workers who routinely spend hours a day interacting with various AI assistants, banging out prompts on a keyboard can quickly become a chore.
“Whether it’s a coding task, helping write a document or think about strategy — there’s just so much typing and typing and typing you do as a part of that,” said Chris Patalano, chief technology officer at Thumbtack, an online marketplace for professional services.
With that in mind, Patalano and other senior colleagues last year began experimenting with new ways to interact with AI systems within Thumbtack. The idea was to test AI-assisted dictation tools developed by startups such as Monologue, Superwhisper, Willow Voice, and Wispr.
Unlike previous generations of dictation apps that aimed to produce a verbatim transcript, newer tools rely on large language models (LLMs) to craft polished, edited text. The companies behind them claim users can produce text several times faster than typing, with greater accuracy than voice tools built into other apps.
That has sparked renewed interest in using voice prompts to carry out routine tasks in the workplace.
After one of Thumbtack’s principal engineers suggested Wispr Flow, Patalano kicked off a small pilot project with a handful of colleagues over a couple of months. The pilot was a success, and Wispr Flow is now available to more than 200 IT and engineering staffers; they use it for a variety of tasks, including interactions with AI assistants and drafting Slack messages to colleagues.
Although Patalano said he still prefers typing for certain apps, Wispr Flow’s AI dictation tool has become a part of his daily workflow. “It’s becoming the primary interface that I have for any AI tools. It’s just so much more effective and efficient than having to type,” he said.
“I’ve used it to help me build prototypes, explore the code base, help me explore my own technical strategy. I’ve used it to do analytics across data sets — even very specific acute things, like ‘What do I need to make sure is on my to-do list this week?’”
A new generation of dictation tools
Software that translates spoken words into text isn’t new to the workplace. Speech-to-text dictation tools have been around in various forms for decades. The earliest example dates back to 1952, when Bell Labs created Audrey, widely regarded as the first automatic speech recognition system. (Audrey could recognize the spoken digits 0-9 with 90% accuracy when used by the machine’s developer, HK Davis.)
Commercial products appeared in the 1980s, with broader adoption in the 1990s via software such as Dragon Dictate. These were specialized — and expensive — applications with limited functionality, appealing mostly to professionals for whom dictation was already a part of their workflow, such as doctors and lawyers, rather than a wide range of office workers.
In recent years, speech-to-text software has become more accessible, especially with the integration of speech recognition into smartphones and computers by Apple, Google, Microsoft, and others. Deep learning has also significantly improved accuracy.
That’s made voice input more common in the workplace and an important accessibility tool for people who find typing difficult — even though the systems can still be “quite brittle,” said Benjamin Cowan, professor at the School of Information and Communication Studies at University College Dublin. That’s especially true of early voice input technology.
“Not only did they get things wrong all the time, they wrote everything you said — even if you didn’t want it to,” he said. “This meant that a lot of time was taken editing the notes after they were dictated.”
Now, several startups offering AI dictation tools, including Wispr, aim to make voice a viable alternative to typing for everyday computer tasks. The key difference from earlier iterations of tools is the use of AI models to edit text in near-real-time, removing disfluencies such as “umms,” “ahhs” and filler words to create a polished sentence.
Whispr Flow offers shortcuts, or “snippets” with its voice tool.
Whispr Flow snippets
In most cases, users can invoke an AI dictation tool across mobile and desktop applications with a text field – whether that’s a document editor, email client, a vibe-coding app or anything else – by pressing and holding a designated key or button while talking. Users can add words to the app’s dictionary so it can pick up on uncommon names, abbreviations, and industry jargon.
“The technology itself has improved dramatically” compared to previous tools that sought to transcribe speech verbatim and could be frustratingly inaccurate, said Maria Bell, senior research analyst at CCS Insight.
“These modern systems are much more contextual; they understand your intent, they can help structure your thoughts and rewrite while you speak. They function more like writing assistants rather than just dictation.”
Wispr is among the best-funded startups in the market, having raised $81 million to date. Bloomberg reported in May that the company was in talks to raise a further $260 million at a $2 billion valuation. Other vendors have also attracted investor backing, with Willow Voice announcing a $4.2 million funding round last year.
The software is typically available via a freemium model, with a free tier offering basic functionality and usage limits alongside paid premium versions. Superwhisper Pro is $8.49 per user each month; Willow Voice’s Team Pro and Individual Pro are $10 and $12 per user each month, respectively; and Wispr Flow Pro costs $12 per user each month. Enterprise pricing is not publicly available from these vendors.
Larger tech firms have also invested in AI-assisted voice functionality. Apple, for instance, recently announced AI-powered dictation for its revamped Siri AI assistant, while Google is building in similar functionality for the Gboard keyboard on Android devices. Google is also developing a standalone AI dictation tool – Edge Eloquent – although its approach differs from that of startups in the space because the tool is not available across separate applications.
Why use AI dictation?
The key promise of AI dictation is that it can increase a knowledge worker’s words-per-minute (wpm) output versus typing.
According to Superwhisper, most office workers can knock out between 40 and 70 words a minute on a keyboard, though some can be much faster. (New York Times reporters vary from 36 to 134 wpm, according to a Times article earlier this year.) People talk much faster, at a rate of 160 to 180 wpm, and AI dictation app vendors promise low latency processing to turn speech into edited text (usually less than a second; some claim under 200 milliseconds).
It’s not just about speed: Willow Voice, for instance, claims its app is three times more accurate than dictation tools built into other applications.
The prospect of accelerating routine writing and communication tasks has obvious appeal, particularly as AI threatens to increase rather than reduce the burden on office workers. “We all feel like we’re working faster – we have to do more with less time,” said Bell.
“Employees are overloaded with communication work, and they’re spending huge amounts of time every day writing emails, messaging colleagues, using generative AI,” she said. “Voice tools are appealing because some feel they can do wor] faster. It reduces friction around all the tasks they’re being asked to do.”
The technology is potentially suited to a variety of jobs, said Cowan — not only those that require dictation — helping with tasks such as writing to-do lists and documents, or sending messages and emails.
Accessibility is important, too. “These dictation tools also mean that people who find it hard to type or cannot type now have much better apps to help them with writing,” said Cowan.
What’s holding the technology back?
Despite these potential benefits, the idea of talking to a laptop or smartphone throughout the day might not be appealing for a lot of people, particularly those in a busy office.
“Some might find it embarrassing or uncomfortable, they’ll be worried about distracting colleagues or creating a disruption,” said Bell. “That’s still a major behavioral barrier that you have to overcome.
“The technology is ready, but maybe workplace etiquette and culture is not necessarily there yet,” she said.
Working remotely, Patalano said he and his team can side-step some of this awkwardness. But it still took time to adjust to voice inputs.
“Because we’re fully remote, we don’t have the challenge of everybody sitting side by side in an office talking into their computers, which would be more challenging, I suspect. But even getting comfortable with talking out loud alone in a room took a minute,” he said.
As with any AI tool, there’s also the question of accuracy.
Even if vendors promise a low error rate, LLM outputs can still have errors, requiring users to check the results. “They can still mis-recognize what’s being said,” said Cowan. Those in high-risk sectors such as healthcare still need to go through the AI-edited text and “double- and triple-check” the dictation.
This friction means extra steps for a user working with the technology.
It doesn’t take much to dissuade workers from adopting a new tool, said Jon Arnold, research analyst at J Arnold & Associates. “There’s definitely a lot of use cases where it would have a lot of value, but you’ve got to trust it — if it’s not giving what you think it will, you’re either going to fine tune it or go back to the keyboard and do it the old-fashioned way,” he said.
There are also privacy concerns. Because some tools send voice data to the cloud for processing, organizations in heavily regulated industries such as finance, healthcare and government might move cautiously.
Bell points to two types of privacy: social, such as “having colleagues overhear what you’re saying,” and digital privacy, which relates to who else can access the conversation data.
App providers take different approaches; some process voice data on device, others send it to the cloud. That’s an important distinction for organizations with strict data protection requirements, said Bell.
“Where’s the voice data processed? Where is it stored? How can it be accessed? Enterprises are very, very focused on governance and data security and data privacy,” she said.
Evan Yang
Too soon to ditch the keyboard?
Despite growing interest in the technology, it’s still unclear whether a large number of workers will choose talking over typing. And remains to be seen whether startups that offer a best-of-breed AI dictation app can gain traction, or fade if the technology simply becomes embedded within the software ecosystems of larger tech firms.
Workers are more familiar with voice technology, thanks to AI assistants in smartphones and smart speakers at home. That, said Bell, could improve the prospects of wider use in business settings.
“Voice interaction feels less niche than it did about five years ago,” she said. “Overall, the technology is improving quickly…, but how we’re really going to determine success is whether we can change human behavior.”
Arnold is bullish about the use of voice technology in the workplace: “Five or 10 years [from now], we won’t think twice about it. It’ll just be the norm.”
Bell is more cautious.She sees potential for AI dictation as a supplementary tool for communication-heavy work. “I don’t think it’s going to replace the keyboard, but I do think it could become a secondary interface,” she said.
Even Patalano doesn’t expect AI-assisted voice dictation to entirely replace typing “Your speaking voice and your written voice will always, to some degree, be different, and that’s okay: we should probably lean into that,” he said.
“I think there will always be a place for wordsmithing, crafting, writing – and the same with coding, too. There’s going to be lots of cases where every single word matters.”
He plans to continue using AI dictation, whether with Wispr Flow or other similar tools that might emerge in the future.
While a lack of accuracy slowed adoption in the past, continued advances could open the door to wider workplace uptake.
“When I try to use a voice tool and it misses even once, you kind of throw up your hands and walk away, because the cost of having to correct it is way more than the benefit of using it versus typing,” Patalano said. “But, especially with the improvements in LLMs and AI models generally, the accuracy of these is going to keep getting better and better.
“I’m already looking for more and more opportunities to use voice instead of having to type.” Talk, talk, talk: The rise of AI dictation tools at work – ComputerworldRead More