Sam Udotong said he and Fireflies.ai cofounder Krish Ramineni manually took notes for over 100 meetings while pretending to be an AI bot called Fred.
The AI startup’s story, posted to LinkedIn by Udotong, raised eyebrows over its “fake it till you make it” approach.
Udotong, CTO of Fireflies.ai, which built a product to automate note-taking for online meetings, wrote that the company’s first batch of customers in 2017 were getting a human-run Fred.
The post soon went viral on LinkedIn, with nearly 3,000 reactions and hundreds of comments. While some applauded the founders for their approach, others raised questions.
Fireflies.ai is now valued at $1 billion, largely thanks to the rise of virtual meetings during the pandemic.
When Ramineni and Udotong came up with the concept of a meeting notetaker AI, their backs were “against the wall,” Ramineni told Business Insider.
“So we said, we’re down to our last bit of money, let’s figure something out, but this time, before we write a line of code, let’s make sure that we can actually validate it,” said Ramineni, the CEO of Fireflies.ai. “And we had to pay rent, and SF is really expensive.”
Ramineni was living off savings after a brief stint with Microsoft, and Udotong had never had a full time job. To validate their idea, Ramineni said, they reached out to some friends in the tech space and asked them if they would be willing to pay $100 a month for their meeting notes to be fully taken care of in 2017.
Ramineni said that he told those friends that there would be some human involvement and oversight, but did not specify that the process was actually entirely manual. Posing as an AI bot called Fred, the two founders said they joined meetings and took notes by hand, and they usually delivered the notes within a day.
“We had just enough money to pay for the rent where Sam was staying, and we found incredible demand,” Ramineni told Business Insider.
Unlike an AI bot, the founders are humans who could not be omnipresent, and Ramineni said it took about a hundred such sessions for them to grow weary of back-to-back meetings and feel the constant stress about being double-booked.
Ramineni said that Fireflies.ai had not approached any investors until they were already working on the fully automated product, and there was no overlap between investment and the founders pretending to be AI.
By late 2018, the product development was full steam ahead, and they had stopped manually taking notes for customers. Ramineni said they were kept afloat by small checks from angel investors. At the end of 2019, Fireflies.ai was able to beta test its product and do live demos for institutional investors, allowing them to raise a seed fund of more than $4 million.
“We told them, actually, for the first couple of users, we sat down, took notes for them, and it helped us realize and understand what it means to do good notes,” said Ramineni of the investors. “And then they were impressed with how we validated the problem first, and then we went and built the product.”
The ‘fake it till you make it’ approach
Tim Weiss, a professor of management and entrepreneurship at Imperial College London, told Business Insider that Fireflies.ai’s approach sounds like “pretotyping,” which is a very common but “questionable” practice.
“Basically, you pretend you have a product to learn about how people engage with it before you build it,” said Weiss. “This is something that is done at the very early stages of a startup to validate an idea, but of course not later on.”
In other words, Fireflies.ai may have simply said the quiet part out loud.
The startup may have skimped on details with early customers, but it did not approach institutional investors until the company had a functioning product to show, the founders said.
Kevin Werbach, professor of legal studies and business ethics at the Wharton School of Business, told Business Insider that, despite the risks, “fake it till you make it” is a “hallowed element” of tech startup culture.
“When done right, it’s Steve Jobs’ ‘reality distortion field’ and countless entrepreneurs that fudged on details to get the time or resources needed to make those claims real,” said Werbach. “When done wrong, it’s Elizabeth Holmes going to jail.”
“Most situations are in the middle,” Werbach added.
The “reality distortion field” is a term coined by Steve Jobs, the cofounder of Apple, referring to his ability to convince himself and others that seemingly impossible tasks are achievable through the force of will.
According to the “Internet History Podcast,” a show that documents the rise of technology that may seem commonplace right now, the iPhone was frequently failing when Jobs made a sleek demonstration of it back in 2007. Engineers determined a specific order of demo actions for Jobs to perform to prevent the phone from crashing, and hard-coded all the demo units to display five bars of cell strength.
Facebook’s stand-alone personal assistant, called “M,” which was shut down in 2018, was also powered by humans to answer the most complex queries, according to The Verge. The goal was that those humans would help train the AI. The product never made it past its private beta stage, The Verge reported.
Ramineni said that as of today, the AI in Fireflies has done over 2 billion meeting minutes and has taken notes for 20 million people, which would be about 4,000 years of meetings if calculated by an eight-hour workday.
“It’s fair to have a lot of skepticism around AI, and I definitely do believe that you have to be transparent in that if you’re raising funds, talking to investors, building the product,” said Ramineni. “You can’t fake it till you make it forever — that’s not how it works.”