CHAPTERS
- 1:00 – 2:20
Recall.ai today: API for meeting recording powering 1,000+ companies
David explains Recall.ai as an infrastructure API that lets developers capture real-time audio/video and other conversation data across Zoom, Teams, Google Meet, phone calls, and in-person sources. He frames Recall as a foundational layer behind 1,000+ AI and SaaS companies, with strong revenue efficiency and rapid growth.
- •API to capture conversation data from multiple sources (video meetings, phone, in-person)
- •Infrastructure layer used by 1,000+ companies
- •Near $20M revenue with <30 people
- •Raw usage grew ~10x in the last 18 months with expectation of another 10x
- •Positioned as “picks-and-shovels” for the AI app ecosystem
- 2:20 – 4:00
Engineering at extreme scale: reliability, EC2 sprawl, and video throughput
The conversation zooms into the operational reality of running conversation-capture infrastructure at high volume. David shares concrete scale metrics and highlights why reliability is existential when recordings can’t be recovered if missed.
- •~8 million EC2 instances launched per month
- •Peak processing around 3 TB/s of raw video
- •Reliability is critical because missed recordings are irretrievable
- •Scale is larger than nearly all companies on AWS (per David)
- •This scale creates a compounding engineering advantage for Recall
- 4:00 – 6:00
Hackathon to YC: a 19-year-old’s sprint from Waterloo to Winter ’20
David recounts winning a YC hackathon in 2019, earning a YC interview, and rapidly reshaping the project into something fundable. He describes the intensity of balancing school, travel, and the all-in commitment required to seize the YC opportunity.
- •Flew to Mountain View using saved internship money to attend YC hackathon
- •Won hackathon; Michael Seibel grilled viability during presentation
- •Had ~3 weeks to make the project “YC interview-ready”
- •Tried selling $20/month licenses across campus to show traction
- •Skipped multiple Waterloo exams to attend in-person YC interview; got into W20
- 6:00 – 8:00
Co-founder transitions and committing through uncertainty (finding Amanda)
David explains why early co-founder setups changed—first due to reluctance to leave college, later due to COVID-era risk and desire for stability. He details how alignment on mission and resilience became the deciding factor in bringing Amanda on as the long-term co-founder.
- •First co-founder didn’t want to leave college; amicable split
- •Second co-founder left amid COVID to pursue a more stable path
- •David spent several months as a solo founder
- •Amanda initially helped with early product materials, design, and website
- •Key alignment: expectations that “things going wrong” is the startup default; commitment to persist until it works
- 8:00 – 10:00
Building the first product: a call recorder before LLMs
Recall’s roots were a standalone call recording product built to manage the team’s own user research and recordings. David describes how the product forced them to become experts in the hardest part—capturing and storing conversations reliably.
- •Built a call recorder starting in 2020 (pre-LLMs)
- •Motivation: massive volume of research notes/recordings was hard to work with
- •Ran the recorder product for nearly three years
- •70–80% of engineering effort went into recording infrastructure
- •Customer expectations made uptime and durability non-negotiable
- 10:00 – 12:00
Pivot to Recall: selling conversation-capture infrastructure as an API
As LLM capabilities accelerated, David and Amanda recognized conversation data was becoming the substrate for a new wave of AI products. They repackaged their hardened recording stack into an API and repositioned from an app to infrastructure.
- •Observed rapid LLM progress enabling new products on unstructured conversation data
- •Insight: many companies would need the same capture infrastructure
- •Pivoted in February 2022 to infrastructure/API model
- •Moved from “end product” (recorder) to “platform” (developer API)
- •Thesis: companies should focus on differentiation, not rebuilding conversation plumbing
- 12:00 – 14:30
Fundraising the hard way: deferred Demo Day, 160K first round, then 120 seed meetings
David details unusually grind-heavy fundraising cycles, first for the recorder and then for Recall. With minimal network and remote constraints, they relied on relentless outreach to founders for introductions, turning persistence into a core operating advantage.
- •Deferred Demo Day; spent ~4 months raising for the call recorder and raised ~$160K
- •Seed for Recall took ~8–9 months in total
- •Spoke with 120+ investors to close the seed (~$2.5M)
- •Sent 1,000+ handwritten emails to founders to get intros
- •Early investors came from a small number of hard-won “yes” decisions
- 14:30 – 17:00
Deep conviction from living the pain: on-call nightmares and infrastructure truth
Asked what proved they were on the right track, David points to firsthand experience operating the system under real customer pressure. The “lived problem” created conviction that outlasted skepticism from outsiders.
- •David personally ran production infra and handled escalations for years
- •Nightmares about getting paged underscored the operational burden
- •Belief: infrastructure pain is real even if investors dismiss it
- •Reliability demands create a high barrier to entry
- •Founder conviction came from direct exposure, not market narratives
- 17:00 – 19:30
Landing first customers by converting competitors into buyers
Recall’s first go-to-market move was counterintuitive: they approached their former competitors. A small number trusted them, adopted the API, and gained speed—creating a visible advantage that helped the category become “standard.”
- •Emailed old competitors immediately after pivot
- •Most were skeptical; a few adopted and offloaded infra burden
- •Adopters moved faster, creating competitive pressure on others
- •Network effects via industry observation accelerated adoption
- •Within 1–2 years, using Recall-like infrastructure became normalized
- 19:30 – 22:00
Learning sales from scratch: $2M outbound and reframing what ‘sales’ is
David explains how the team learned enterprise sales without prior experience, largely through repetition and direct founder-led outbound. He demystifies sales as communicating value and navigating internal buyer processes—not manipulation.
- •Worked with major customers (e.g., HubSpot, ClickUp, Monday.com)
- •First ~$2M in revenue came from pure outbound led by Amanda
- •Sales = (1) clearly communicating value, (2) helping customers execute procurement/adoption
- •Engineers often mischaracterize sales as “black magic” or manipulative
- •Founder-led selling delayed the need to hire a sales org early
- 22:00 – 24:00
Reaching ~$20M ARR with <30 people: high bar hiring and “full-stack” ownership
The discussion turns to extreme revenue-per-employee and how Recall stays lean intentionally. David attributes it to hiring rare, high-agency operators who span engineering, product, and customer interaction, minimizing information loss from handoffs.
- •Lean team is enabled by a very high talent bar, not just frugality
- •Engineers own product decisions and talk to customers/sales
- •Avoiding handoffs preserves nuance; each handoff loses ~90% of info (as cited)
- •More than half of engineers are former startup founders
- •High-agency culture reduces need for heavy process and middle layers
- 24:00 – 26:00
Why conversation data is the future of AI: the ‘missing context’ inside companies
David lays out the macro thesis: spoken language at work dwarfs written artifacts, and most organizational context never enters docs. For AI to be effective, it needs access to conversations—the living substrate of how decisions and knowledge move.
- •Claim: ~5x more words are spoken at work yearly than exist on the internet
- •~99% of company context is not written down; it’s exchanged verbally
- •AI can’t become effective by reading docs alone; it needs conversational context
- •Recall becomes a foundational data layer for AI applications
- •The platform benefits from visibility into emerging AI use cases across industries
- 26:00 – 28:00
What’s next: botless desktop recording, broader capture surfaces, and data infrastructure
David previews product expansion beyond meeting bots, including a desktop recording SDK and future support for phone, mobile, storage, querying, and preprocessing. The goal is to widen capture coverage and provide end-to-end primitives for working with conversation data.
- •Launched a desktop recording SDK to capture without a meeting bot
- •Captures meetings from the user’s machine and supports in-room/ambient recording
- •Roadmap includes phone call recording and mobile recording
- •Plans to offer storage, querying, and preprocessing for conversation data
- •Mission: let customers focus on differentiation while Recall handles heavy infrastructure
- 28:00
Hiring and founder advice: high-agency teams and ‘don’t give up’ for at least a year
Closing themes emphasize hiring people motivated by ownership and intensity, given Recall’s role as critical infrastructure for much larger companies. David’s founder advice centers on endurance: competence and product clarity compound over time, and most quit before reaching that point.
- •Hiring across engineering, product, marketing, and sales
- •Seeking high-agency people who want to push themselves and own outcomes
- •Operating as critical infrastructure raises the standards and responsibility
- •Advice: don’t give up—success requires learning many skills over time
- •Expectation-setting: it can take a year or more to know if a product will work
