The origin of Recall.ai and the problem nobody wanted to solve
Recall.ai came out of a scenario that many startup founders know all too well: the frustration of spending most of your time solving problems that have nothing to do with your core product. In 2020, David Gu and Amanda Zhu were building a video conferencing recording tool aimed at product managers and UX researchers. The idea was great and demand was growing, especially during the pandemic, when recording a Zoom call was still something new and even socially awkward for a lot of people. But in practice, about 80% of the team’s effort went into keeping the most tedious and complex part of the system running — the infrastructure for connecting to platforms like Zoom, Google Meet, and Microsoft Teams.
Every update from those tools broke something, every API change required manual fixes, and the product they actually wanted to ship was always on the back burner. David described the situation as a dead end: if the infrastructure worked perfectly, customers didn’t even notice and just asked for new features. But if something went wrong and a recording was lost, the consequence was losing the customer. There was no way to win that game. That’s when it clicked: if they were struggling this much, probably hundreds of other companies were too.
The strategic pivot happened when the pair decided to stop fighting the problem and turn it into the business itself. Instead of continuing to build a recording end product, they chose to build the infrastructure layer that any company would need to access conversation data from online meetings. The idea was simple on the surface but incredibly hard to pull off in practice: create a unified API that would let any developer connect their product to virtual meetings, capture audio and video in real time, and process that information without having to deal with the technical complexity of each platform individually. In other words, Recall.ai took on the heavy lifting so other companies could focus on what actually matters — building smart, useful AI products.
This decision wasn’t trivial. Walking away from a product that already existed to bet on an invisible layer of technology takes guts and a very precise read of the market. But David and Amanda had an important edge: they were probably two of the five or ten most experienced people in the world in this specific domain of reliable meeting recording infrastructure. All the expertise they’d built up while struggling with the previous problem became the foundation of the new business.
The turning point: when LLMs changed everything
Recall.ai’s timing wasn’t a fluke. In 2022, when language models like GPT-3.5 started showing an impressive ability to process unstructured text, the market shifted in a radical way. Suddenly, two-person startups were building products that would’ve been impossible just two years earlier. Automated meeting summary tools, sales assistants that analyzed conversations in real time, coaching platforms — all of it started popping up at a breakneck pace.
David noticed something fundamental at that moment: LLMs solved the smart part of the equation, but they didn’t touch the infrastructure problem. In other words, even with increasingly powerful models, every company that wanted to build a product based on conversation data still had to solve the same technical nightmare that he and Amanda had already faced. Demand for reliable infrastructure to capture conversations was about to explode, and the infrastructure-as-a-service market for this type of data simply didn’t exist.
With that clear read on the opportunity, the pair took the infrastructure they’d already started building for the previous product and began extracting it into a standalone service. Before they even had a complete product to sell, David and Amanda did something bold: they sent cold emails to their former competitors in the meeting recording space, offering to build the infrastructure for them. The responses were telling. Some ignored them, others were suspicious. But several replied right away with genuine relief, basically saying they were thrilled that someone was finally solving this incredibly painful problem. That early validation was the fuel they needed to move forward with conviction.
How Recall.ai’s API works in practice
The technical pitch behind Recall.ai is offering an API that abstracts away all the complexity of connecting applications to virtual meetings. In practice, that means a developer can, with just a few lines of code, send a bot to join a meeting on Zoom, Google Meet, Microsoft Teams, or Webex. That bot captures the conversation audio, identifies who’s speaking, generates real-time transcriptions, and delivers everything in a structured format to the application that made the call.
Receive the best innovation content in your email.
All the news, tips, trends, and resources you're looking for, delivered to your inbox.
By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.
It sounds straightforward when you describe it that way, but under the hood there’s a robust infrastructure that has to deal with massive technical challenges. David explained one of the most complex ones pretty clearly: meetings don’t happen randomly — they cluster at specific times. Every day at 9 AM Pacific, millions of people on the West Coast jump into their daily standups at practically the same time. It’s like a Black Friday traffic spike, except it happens every single day. The infrastructure needs to scale instantly, process data in real time, and tolerate zero data loss. A single failed recording can mean a lost customer.
One of the most important differentiators is the ability to deliver conversation data in real time, not just after the meeting ends. This opens up huge possibilities for AI products that need to act during the conversation — like assistants that suggest live responses for sales teams, tools that detect customer objections the moment they happen, or compliance systems that monitor regulatory calls while they’re still in progress. Recall.ai’s API also offers features like speaker detection, precise timestamps, and metadata that make indexing and searching later much easier. All of it delivered in a standardized format, regardless of which video conferencing platform is being used, which eliminates the headache of maintaining separate integrations for each service.
The architecture was designed for scale from day one. Recall.ai’s infrastructure supports everything from startups processing a few dozen meetings a week to enterprise companies with thousands of simultaneous calls. This flexibility is possible thanks to a bot orchestration system that manages resources dynamically, spinning instances up and down based on demand. For developers, the experience is seamless — they make a call to the API, get structured conversation data back, and don’t have to worry about anything happening behind the scenes.
The product’s evolution in three waves
Since its founding in 2022, Recall.ai has expanded its portfolio strategically, following a natural evolution driven by market needs. The product has evolved in three major waves:
- Meeting Bot API: this was the foundation of everything — the core API for video conferencing recording. This layer solved the immediate problem that companies like David and Amanda’s former startup faced, offering a reliable and scalable way to send bots into meetings and capture data.
- Desktop Recording SDK: launched in late 2025, this development kit enables AI-powered note-taking-style recording that can be embedded directly into any desktop application. Within just a few months of launch, the company had already racked up hundreds of thousands of installations and deployments.
- Mobile Recording SDK: the next frontier, which will extend the infrastructure to phone calls and in-person meetings, significantly broadening the platform’s reach beyond the video conferencing world.
This progression shows a clear long-term vision. Recall.ai isn’t just solving the problem of recording Zoom meetings — it’s building a complete platform for capturing conversation data in any context, whether online, over the phone, or face to face. As AI agents become more autonomous and actively participate in different types of interactions, this breadth becomes increasingly valuable.
The usage-based pricing model
One aspect that stands out in Recall.ai’s strategy is its pricing model. Instead of charging for licenses or number of users — like most traditional SaaS companies do — Recall.ai adopted a consumption-based model, charging per hour of conversation data captured. This approach creates what David calls success alignment between the company and its customers.
In practice, it works like this: if a customer’s product is doing well and processing large volumes of data, Recall.ai naturally earns more. But if the customer is still in testing mode, working with a small pilot group, the cost is proportionally lower. This eliminates several common problems with traditional models:
- No surprise bills, because costs only go up as the customer’s product generates more value
- No artificial per-seat pricing friction that can limit adoption within an organization
- No need for complex plan structures and tiers, because usage naturally segments customers by scale
This model mirrors the pricing logic that developers already know from services like AWS, OpenAI, and Anthropic. Predictability and elasticity go hand in hand — customers can estimate costs based on expected volume, while pricing adjusts automatically as that volume grows or shrinks. The model works for both seven-figure contracts and customers who opt for self-serve with pay-as-you-go.
The counterintuitive sales strategy
Here’s one of the most interesting parts of the Recall.ai story, and it goes against pretty much everything you’re taught about how to scale developer products. Conventional wisdom says that dev tools need to be self-serve from day one, with flawless documentation and bottom-up adoption. David and Amanda did the exact opposite.
For the first three years, every customer who wanted to use Recall.ai had to talk to someone on the team. In the early months, that someone was David or Amanda themselves. The logic behind this decision was simple but powerful: the real value of those interactions wasn’t the revenue — it was the learning. Every conversation with a potential customer followed a consistent set of questions — what are you trying to build, why are you building it, why do your customers want this solution, and how can Recall.ai help.
Across dozens of verticals — telehealth, financial services, sales, customer success — David and Amanda built a deep understanding of the market that would’ve been impossible to get from analytics dashboards or support tickets alone. They learned about regulatory constraints in healthcare, understood why some industries needed phone call recording instead of video, and observed that different end-user profiles had varying levels of comfort with recording technology.
That investment in learning started paying concrete dividends when, in early 2025, Recall.ai finally launched its self-serve model. With all the market intelligence accumulated over three years of direct conversations, the product arrived already calibrated to meet developers’ real needs. Today, thousands of companies sign up on the platform every week.
The AI product ecosystem that depends on this infrastructure
When you look at the current landscape of AI products focused on productivity and enterprise communication, it’s striking how many of them depend, directly or indirectly, on the infrastructure layer that Recall.ai provides. Companies like HubSpot, ClickUp, monday.com, and PagerDuty use the platform to power features ranging from automated meeting summaries to advanced team performance analytics. In many cases, the end user has no idea Recall.ai even exists — and that’s exactly the point. The company operates as an invisible layer, supplying the conversation data that fuels increasingly sophisticated AI experiences.
This strategic position in the ecosystem creates a competitive advantage that’s tough to replicate. As more companies build AI products that need access to conversations — and this trend is only accelerating with the evolution of language models — demand for Recall.ai’s API grows naturally. Every new use case, whether it’s an AI agent that joins meetings to take notes, a system that analyzes the emotional tone of support calls, or a training tool that evaluates sales rep performance in real time, represents a potential customer that would rather pay for a ready-made solution than invest months of engineering to build something equivalent from scratch.
Today, more than 3,000 companies use Recall.ai to power their products. The network effect here is powerful: the more companies that use the platform, the more robust and reliable it becomes, which attracts even more customers. Recall.ai’s client list serves as validation that some of the biggest and most innovative software companies in the world see conversation data as critically important to the future of their products.
Hiring and culture: engineers who understand the business
Building an infrastructure company from a two-person team brought unique challenges, but also unexpected advantages. David pointed out that Recall.ai’s hiring model has always sought a specific type of engineer — someone with strong technical depth in backend and infrastructure, but also with enough product mindset to talk directly with customers and understand what they need.
David admits that these two traits are, in a way, anticorrelated in the job market. It’s rare to find engineers who are equally strong on both the technical side and the business understanding side. But when you do find that profile, the result is an extremely efficient operating model where engineers don’t just understand how the software works — they also understand how the business works and what customers expect.
This hiring philosophy reinforces something David mentioned about the importance of understanding the entire chain of causality — from the infrastructure all the way to the value that reaches the end user. Often, Recall.ai’s customers don’t even know exactly what’s good or bad about the technical implementation, because they’re waiting for their own customers to tell them. Having engineers who understand this full dynamic allows Recall.ai to not only provide infrastructure but also guide its customers on best practices, cost optimizations, and product strategies — something that’s only possible when the technical team has that holistic understanding.
Lessons for anyone building AI products today
The Recall.ai journey offers valuable insights that go well beyond the meeting recording niche. For anyone building in the AI space, a few lessons stand out:
- Domain expertise creates advantages that compound over time. David and Amanda were among the most knowledgeable people in the world about meeting infrastructure when they started. Understanding both the developer and the end user allowed them to predict cost structures, suggest optimizations, and advise on product strategy — all based on lived experience.
- The most visible product isn’t always the most valuable one. Recall.ai didn’t win by racing to build the flashiest AI feature. It won by solving the complex, unglamorous problem that every AI product needs to solve but nobody wants to touch.
- Hands-on sales can work for developer products when the goal is learning. The decision to force conversations with every customer during the first three years built a market understanding that couldn’t be replicated with analytics or support tickets.
- Infrastructure pricing should align the vendor’s success with the customer’s. When your costs scale alongside your customer’s success, pricing becomes transparent and predictable for both sides.
Conversation data as the missing piece in AI automation
Perhaps the most thought-provoking takeaway David shares is about the future of AI agents and enterprise automation. Think about this scenario: imagine onboarding a new employee at your company using only written information — emails, documents, manuals — and sending them off to work without having them sit in on a single conversation. Even if that person were the smartest in the world, they’d be missing context. Cutting them out of meetings and conversations would leave a massive information gap.
That’s exactly what’s happening with AI agents today. Even with increasingly sophisticated models, better frameworks for agentic workflows, and more elegant interfaces, an agent that needs to draft a sales follow-up email simply doesn’t know what was discussed in the meeting — and the automation falls short of its promise. Without access to the conversations where most corporate context actually lives, even the best-designed agent is going to struggle to execute tasks with the same ease as a human.
As AI agents become more present across every industry, data becomes the foundation that determines whether automation actually works. Companies that figure out how to capture, process, and leverage conversation data efficiently will build AI products that feel genuinely useful instead of limited. Recall.ai positioned itself early in this space, providing the infrastructure that connects AI models to the real conversations where decisions happen. And with the arrival of increasingly autonomous agents capable of actively participating in meetings, negotiating, answering questions, and making decisions in real time, the need for this infrastructure layer is only going to grow 🚀
