Fireflies CEO walks back CTO's viral claim that founders faked early AI

The fake-it-till-you-make-it nightmare

Imagine paying $100 a month for cutting-edge artificial intelligence, only to realize the "algorithm" silently taking notes in your private executive meeting was actually two tired founders surviving on cold pizza.

That awkward reality just blew up on LinkedIn this week. Sam Udotong, the CTO and co-founder of popular meeting assistant Fireflies, posted a boastful retrospective revealing that their transcription service wasn't automated at all in the beginning. Instead, Udotong and CEO Krish Ramineni secretly dialed into early customers' conference calls and typed up the meeting notes completely by hand.

Instead of applauding startup hustle, the internet recoiled. Commenters blasted the post, pointing out blatant privacy violations, corporate data risks, and straight-up customer deception.

The cleanup and the unicorn math

Fireflies CEO Krish Ramineni quickly went into crisis mode, effectively throwing his co-founder under the bus when pressed for answers by Inc.

Ramineni claimed they merely surveyed ten friends about what they would pay for an executive assistant service pairing AI with humans in the loop. He then contradicted Udotong's story entirely, insisting that the Fireflies platform operated from day one without human reviewers, without humans in the loop, and without third-party data labelers.

Adding to the sketchiness, the company's financials don't look completely clean either. Fireflies claimed a $1 billion valuation over the summer following a secondary tender offer allowing employees to cash out. Yet records on PitchBook showed its last formal valuation parked near $72 million, and its recent investor F7 Ventures typically writes checks averaging only $3.7 million. Primary rounds, not secondary tender offers, are what officially build real unicorns.

Fireflies is far from alone in running mechanical-turk theater. App builder Builder.ai went belly up this year after investigations revealed its automated app generator was actually powered by armies of engineers in India. Amazon also took heat last year when reports surfaced that its supposedly automated Just Walk Out checkout tech relied on roughly 1,000 workers in India verifying video streams.

Why it matters

There is a massive legal gulf between testing a landing page and having unvetted humans eavesdrop on private business calls.

For enterprise users, this should trigger immediate vendor audits. Teams regularly discuss sensitive roadmaps, proprietary code, and non-disclosure agreements on transcription calls. Believing you are interacting with an isolated, encrypted model is fundamentally different from letting two tech founders listen in unannounced.

In our view, Udotong casually boasting about this on social media reveals a blinding tech-bro hubris. The Silicon Valley trope of building the frontend before the backend works fine for a marketplace, but applying it to confidential enterprise intelligence crosses directly into gross negligence.

This incident will inevitably accelerate enterprise scrutiny of AI startups. Expect corporate legal teams to demand rigorous, auditable proof that zero human eyes touch data streams before signing another six-figure enterprise contract.