“There are companies out there that will burn hundreds of millions of dollars to get to a billion-dollar milestone,” says Krish Ramineni, co-founder and CEO of Fireflies AI. “They’re burning $5 for every $1 that they make. And that’s not the right approach.”
Fireflies has taken a more deliberate path. Since its founding in 2016, the company has grown into a $1 billion business by staying lean, betting early on voice AI, and avoiding the urge to overspend for growth.
In an interview with AIM Media House, CEO Krish Ramineni shared how Fireflies got there, and where it’s headed next.
“Sometimes when you take venture capital when you don’t need it, it can lead to bad habits.”
“We did raise funding previously, our seed and Series A,” Ramineni says. “But we haven’t raised [further] funding in over four years.” That deliberate choice, he adds, allowed Fireflies to stay in control of its burn and avoid the common pitfall of overextending for growth.
Fireflies began long before the rise of ChatGPT or widespread investor enthusiasm for AI. Ramineni and his team were building transcription and note-taking tools at a time when both the cost and accuracy of language models were shaky at best.
“We were early to this market when this category of AI meeting assistants: note takers and meeting transcription, didn’t exist,” he says. “We pioneered the approach where the bot joins [meetings].”
“What we thought was really special, AI is able to figure out and do.”
Still, building early didn’t guarantee success. “2016 to 2018 was like walking in a desert and not having a map,” Ramineni says. “So either you have to persevere all the way through and you’re going to run out of water, or you have to head back and you’re still going to run out either way.”
Despite the risks, the team stayed the course. “We saw around 2018 some early research around deep learning, LLMs, reinforcement learning that made us believe if you can do this stuff in robotics, [or] Go and chess…” he says. “That gave us conviction [that] language will be a solved problem in the next couple years, but, we can’t wait for the next couple of years to come.”
To that end, Fireflies built its own stack and continued to iterate. The product wasn’t perfect out of the gate. “It was good enough, but it wasn’t excellent,” says Ramineni. But early users stuck with it.
That changed in 2022, when Fireflies gained early access to OpenAI’s GPT-3 API, thanks to an introduction from mutual investor Vinod Khosla to Sam Altman. “This stuff started to look magical in terms of the outputs and the summaries,” Ramineni says. “And then you have this continued progression of GPT-4 and all these new models.”
The turning point came in 2023. “Whatever we were pitching in terms of the product versus what it actually delivered, it met people’s expectations. In fact, it exceeded people’s expectations. It felt like magic.”
“Build something that even your customers would be amazed by.”
Fireflies’ go-to-market strategy relied on product virality. “That was the magic’s secret sauce,” Ramineni says. “You’ve seen in the past, companies have done that really well, like Slack and Dropbox… and Fireflies definitely had that.”
The result: Instead of spending on marketing, Fireflies invested in engineering and product. “We put it all back into R&D and building a really great product experience.”
Beyond Notes: Agents and Live Assist
Today, the company’s ambitions extend far beyond note-taking. Fireflies is developing AI agents that can autonomously attend and conduct repetitive meetings: sales inquiries, candidate screenings, basic support calls. The company has already used its AI recruiter tool to run more than 600 screening interviews internally.
“AI that goes to the meeting, not just helps with it-that’s coming this year.”
“We believe those are the sort of meetings that you should save yourself time from and let AI [handle],” he says. “A lot of times those meetings are pretty repetitive.”
For more complex calls, Fireflies is working on a tool tentatively called Live Assist: a real-time AI that can guide users during meetings with prompts and suggestions. “What questions should you ask based on how the conversation is going, or if you need to dive into something more in detail,” says Ramineni. “That, to me, is also really proactive.”
The goal is to make Fireflies useful before, after, and even during a meeting.
Meanwhile, the company is expanding its footprint across platforms. “It’ll work whether it’s on a video call, a Chrome extension, mobile,” Ramineni says. “It works across Zoom, Teams, Google Meet… you can upload audio files, record meetings directly from your browser tab.”
“Note-taking is just the entry point. Fireflies is turning into a platform.”
Why Fireflies Isn’t Chasing Bigger Models
On AI model strategy, Ramineni pushes back on hype around fine-tuning. “For at least our use case in most SaaS applications, it makes little to no difference,” he says. “I’m saying this directly from conversations with [the] folks at OpenAI.”
What matters more, he says, is domain-specific context. “If a doctor is using Fireflies, we want it to accurately pick up the medical terminology and words,” he explains. “If someone in construction is discussing stuff, there are unique words there.”
He also cautions against the default of using massive models. “Sometimes bigger is not always better,” Ramineni says. “There have been studies that showed these large reasoning models, if they’re actually given more time to think, they overthink just like a human does, and actually provide worse answers.”
This reflects Fireflies’ overall approach: focus on what works, run lean, and design AI systems that integrate cleanly into how people already work.
As Fireflies looks to push toward ubiquity across platforms and meeting types, Ramineni sums up the company’s ambition:
“We are really building one AI that works everywhere. You set it up once: and it doesn’t matter what meeting you have, what calendar you use, or whether it’s in person or over a call.”
