When the history of pandemic-era workplace technologies is written, Otter.ai may end up as a footnote – one of many tools that rode the Zoom boom by offering to turn the world’s hastily arranged video calls into neat, searchable transcripts.
But the company’s co-founder and CEO, Sam Liang, thinks he is sitting on something more valuable than a fancy dictation machine.
The company recently crossed $100 million in annual recurring revenue, impressive for a product that many office workers still use on individual accounts.
But Otter’s model is now bumping into its own success.
With employees at Fortune 500 firms already having adopted it quietly, corporate security teams are belatedly discovering that thousands of sensitive internal meetings have been piped into a consumer-grade app.
Liang puts it plainly: “Many enterprises start to reach out to us after they observed hundreds, sometimes even thousands of users are using Otter on their own.”
That dynamic is both an opportunity and a liability. Shadow adoption is what once helped Slack and Dropbox spread rapidly – until enterprises became nervous.
Meeting data is more sensitive than chat logs or file folders: it contains half-formed ideas, off-hand comments, trade secrets and remarks one never intended to preserve for any length of time.
The sudden institutional desire to corral these recordings says less about enthusiasm for AI than about anxiety over compliance and exposure.
Voice As The Last Unstructured Frontier
Liang’s main pitch is that Otter is not merely an AI notetaker but a “voice-first” AI company in a world obsessed with text.
He is right that most large language models were trained on oceans of writing rather than natural speech.
And if, as he predicts, the proportion of AI-generated written content approaches 90 percent, text becomes less an authentic signal of human communication and more a soup of predictive guesses.
Voice, by contrast, remains messier and harder to fake at scale.
“Most of the AI tools today focus on written documents,” he says.
“Very few tools actually focus on voice and meetings. That’s what we focus on.”
He argues that as AI generates more written content, the relative value of spoken communication will rise – not decline.
“People will never stop talking,” he adds.
But building a moat around speech recognition is difficult when the underlying research is becoming commoditised almost as quickly as text.
Still, Otter’s more interesting idea is not transcription at all but aggregation.
A large company might generate millions of hours of meetings a year. Today, these exist as isolated artefacts – a transcript here, a summary there – ignored by everyone except the poor soul tasked with taking minutes.
Otter wants to weld thousands of these fragments together into an institutional memory: a searchable bank of who said what, and when, across departments and time.
“We create this system for you to manage all of this and have AI extract high-level insights – that’s the next stage.”
The promise is seductive: no more decisions lost to the ether; fewer reinventions of the wheel; fewer “didn’t we discuss this already?” debates.
But it also raises questions. If every meeting becomes part of a corporate watchtower, will anyone still speak freely? Will meetings become even more circumspect and formulaic? A perfect memory is as much a curse as a blessing.
The Agent Enters The Room
Otter’s next frontier is not listening but acting. Liang describes a future in which AI agents will be able to answer questions in real time (“Otter, what was our Q3 growth rate?”) and eventually participate as quasi-colleagues.
He frames this as inevitable: AI with infinite memory and instant recall will simply become part of the conversational fabric of work.
Here again, the logic holds – but only up to a point.
AI agents may be able to improve the tempo of information flow, but they also risk compounding errors at machine speed.
Liang insists on a human-in-the-loop for “high-impact” actions, though his definition of “low-impact” – for instance, pushing data automatically into Salesforce – may cause some enterprise security officers to twitch.
More revealing is what Liang did not say: that enterprises might worry about AI quietly influencing decisions, misinterpreting context or entrenching biases that humans do not detect until too late.
Compliance: The Necessary Sedative
To win over healthcare and finance – industries that balk at the idea of any third party listening in – Otter is racing to accumulate every acronym available: SOC 2, GDPR, HIPAA.
Liang describes an increasingly granular privacy architecture: configurable retention schedules for audio, transcripts and summaries; the ability to pause recording mid-meeting, and an upcoming “blackout” tool to scrub sensitive segments.
These are sensible moves, if reactive.
They also underscore the tension at the heart of Otter’s strategy: the more comprehensive the memory, the more elaborate the mechanisms required to forget.
Liang insists that recording meetings will soon be default behaviour in companies, the bigger question is whether employees agree.
The Competitive Squeeze
Liang’s confidence in Otter’s technical depth is unambiguous.
He argues that controlling the whole voice-AI stack – speech recognition, speaker identification, text-to-speech – gives his company a speed and cost advantage that others cannot replicate.
But Fireflies, Fathom and a dozen others have shown that strong user experience, tight integrations and enterprise hand-holding often matter more to buyers than technical purity.
Meanwhile, Microsoft, Google and Zoom can deploy meeting features overnight to hundreds of millions of users – without needing to persuade security departments to approve a new vendor.
Otter may find itself in the same position as Slack: widely admired, often superior, and slowly eaten from the edges by incumbents with distribution.
The Meeting Reimagined – Or Merely Recorded?
Liang’s vision of the future workplace is unmistakably optimistic: meetings infused with instant answers, cross-departmental insight and AI assistants who know when to speak and when to remain silent.
The reality will be messier. AI may accelerate some decisions while distorting others, and it may democratise institutional knowledge or exacerbate surveillance fears.
What is certain is that the meeting is becoming a dataset, whether workers like it or not.
Otter is trying to position itself as the system of record for that data before someone else does.
The question is whether enterprises, having belatedly realised their employees are already feeding countless hours of conversation into an external AI model, will embrace that future or recoil from it.
