Fireflies.ai began its journey as a passive meeting notetaker in the pre-LLM era, but rapid advancements in artificial intelligence have transformed the platform into an active “AI Teammate” capable of executing multi-step workflows across over 100 enterprise tools. In this interview, Krish Ramineni, co-founder and CEO of Fireflies.ai, details the company’s architectural shift from basic transcription to autonomous task execution, particularly through Voice Agents that are saving recruiting and sales teams hundreds of hours.
Ramineni also discusses the critical boundary between AI automation and human judgement, the company’s strict data privacy standards, and its strategic focus on India. As Fireflies.ai’s second-largest market by users, India is actively driving the company’s roadmap, prompting support for 11 regional languages and the development of a localised, lower-cost pricing tier.
1. Fireflies built its brand on transcription and summarisation. What specific architectural or technical shifts were required to move Fireflies from a passive meeting observer to an active ‘AI Teammate’ capable of executing multi-step tasks across external tools?
We started Fireflies in the pre-LLM era, four years before the technology even existed, so the first real shift was getting transcription accurate enough to build anything on top of. The next shift was LLMs. We got early access to the OpenAI APIs almost a year before most people did. That’s what took us from “here’s what was said” to actually understanding it, summarising it, identifying action items, and following context.
Fireflies already have the context across your meetings, your inbox, and the apps you work in, and it uses that to actually do the work: drafting the email, updating the CRM, creating the report. That’s the part that goes beyond understanding; it goes into those tools and gets it done. You don’t need a dozen separate tools for each of those tasks. You need one AI teammate that already has the context to execute them.
2. You are planning to roll out multiple new capabilities over the next year. How will Fireflies break out of the virtual meetings to become an everyday execution layer across broader enterprise workflows?
We already have. Fireflies started as an AI Notetaker because meetings were the easiest place to capture what’s going on. That’s not the only place anymore, it’s in the inbox, in Slack, CRM, and across 100+ tools you use.
Work doesn’t happen in one place. It’s a meeting, then an email, then a Slack thread, then a CRM update, and most software only sees one piece of that, especially across a team: a deal moving through sales, a candidate moving through recruiting, a ticket moving through different owners. Enterprise workflows usually break at the handoff, not inside anyone meeting.
That’s the next phase we’re already in, rolling out products built on the same idea: gather the context so you’re never re-explaining yourself, then actually do the work and take the action for you.
3. Voice Agents have already logged 40,000+ conversations globally. What is the most complex, end-to-end workflow a Voice Agent is currently handling post-call without human intervention?
The two top use cases are screening calls and sales discovery calls. For recruiting: you set up the Voice Agent once. Fireflies run the calls, ask the questions, and follow up based on the candidate’s answers. Right after the call, it prepares a clean summary and a scorecard and pushes the notes into your ATS. Nobody has to write anything up or enter it manually. Our own recruiting team has saved more than 800 hours doing exactly this.
For sales: Fireflies runs the discovery call, qualifies the lead, and structures what it learned into a clean sales-qualification summary. The moment the call ends, that goes straight into the CRM as ready-to-use notes. No rep has to sit down and type it.
4. As AI takes on more operational tasks generated during live conversations, what work should businesses confidently delegate to AI, and where must humans permanently remain in control?
Delegate anything high-volume and low-ambiguity – screening, qualification, follow-through, drafting, notes, routing. If the job is “capture this and get it where it needs to go”, AI should already be doing that, not eventually.
Where I draw the line is judgement calls that actually affect a person, hiring decisions, anything financial, anything where the right answer depends on context a model genuinely can’t see. AI should get you to the decision faster with better information in hand. It shouldn’t be the one making it.
5. India is your second-largest market. How are local enterprise demands driving your engineering roadmap, particularly regarding regional language support and India-specific pricing structures?
India has always been an important market for us, our second largest by users, and we’ve tripled paid teams there over the past two years. When a market shows up that clearly, we build for it. We already support 11 Indian languages, Hindi, Punjabi, Gujarati, Bengali, Malayalam, Tamil, Telugu, Kannada, Assamese, Marathi, and Sindhi, as part of our broader 100+ language coverage, and that list grows every time we see real demand for the next one.
We’re also building a regional plan specifically for India, a lower-cost tier for a market that’s price-sensitive but growing fast. More to share on that soon. Beyond that, there’s no separate “India roadmap”. We listen to every customer everywhere, and India just generates a lot of that signal because it’s such a large share of our base.
6. “How do you plan to leverage India’s AI talent pool to fuel Fireflies’ next growth phase, and what specific technical or product roles are you prioritizing as you expand the local team?”
Fireflies is a fully remote company, with 120+ people working from wherever they are around the world. We hire for talent, not geography. That said, we’ve historically seen a lot of excellent people come out of India, across engineering, product, and non-engineering roles alike. India has an exceptional technology and AI talent pool, and we expect the team here to keep growing as Fireflies expands.
7. Granting AI agents deep contextual access across connected workplace systems naturally increases security risks. How does Fireflies balance proactive autonomy with granular data privacy and enterprise governance?
The more context Fireflies have, the more useful it becomes, but that only works if businesses trust how that context gets handled. Fireflies don’t train AI models on customer data. Customers keep control of their information, and admins control who can access it and how. We also hold ourselves to SOC 2 Type II, GDPR, and HIPAA standards.
But trust isn’t just a certification. It’s also about what you let an AI actually do. Reading something, drafting something, and taking an action are three different levels of authority, and they deserve three different permission levels, not one blanket “yes.” The goal isn’t to give AI as much autonomy as possible. It’s to give it exactly enough to take the busywork off your plate, while you stay in control of the decisions that actually matter.
