By playing, you agree to uListen's Terms and Privacy Policy and YouTube's Terms of Service.
Sourcery with Molly O'SheaInside fal’s $1.65B Series E & F Led by Sequoia & Bessemer
Summaries, chapters, transcripts, and FAQs on this page are made by uListen, not YouTube.
CHAPTERS
- 0:00 – 2:05
Playful name-pronunciation banter, then setting the scene at Dalida
Molly jokes around with Burkay and Görkem about how to pronounce their names correctly before kicking off the episode. They share where they’re filming (Dalida) and a personal story about a wedding there, setting a relaxed tone before talking business.
- •Molly practices and corrects pronunciation of Görkem and Burkay’s names
- •They’re filming at Dalida, an Eastern Mediterranean restaurant with Turkish influence
- •Burkay shares he got married there and Görkem officiated
- •Molly frames the conversation as both "rich" and "delicious" like the food
- 2:05 – 3:11
Why fal stopped chasing fundraising headlines and focused on the business
Molly congratulates the founders on raising multiple rounds quickly and notes they seem less interested in media coverage now. Görkem explains why announcements matter less over time and why they prefer focusing on building the business.
- •Molly describes repeatedly trying to “catch” fal between rounds
- •Görkem says announcements can help early but become less relevant later
- •They prefer focusing on business fundamentals over hype
- •Molly characterizes it as moving past the adrenaline/hype cycle
- 3:11 – 5:00
Generative media as a core AI product-market fit—and why individuals spend so much
Görkem lays out a framework: AI has three big product-market fits, with generative media as one of them alongside search and coding agents. He argues generative media is unusual because individual users can spend extremely high amounts monthly when it’s their job.
- •Görkem’s three AI PMFs: search, coding agents, and generative media
- •He says generative media is one of the only areas where individuals can spend $10k–$20k/month
- •He compares this spending pattern to coding-agent subscriptions
- •Molly asks if they were surprised by the amount of money behind it
- 5:00 – 9:35
The "infinite demand" dynamic: video costs, quality jumps, and revenue mix surprises
The founders describe being surprised multiple times: video quickly became a major share of revenue, then image editing revived image-model demand. Burkay recounts a sharp pricing jump with a video model and how demand still surged as professional use cases expanded.
- •They were surprised by the shift from image to video and the changing revenue split
- •Görkem says image editing brought renewed value to image models and helps video workflows
- •Burkay describes prices settling, then a model debuting at much higher cost—and users still paying
- •They connect higher quality to more professional use cases like ads and production
- •They argue generative media isn’t near diminishing returns yet
- 9:35 – 13:20
Ads as a near-term killer use case—and the feedback loop of performance testing
Molly connects generative video to advertising needs, especially longer ads and performance-driven creative iteration. Görkem explains why UGC-style ads fit current model capabilities and envisions automated improvement loops from performance data; Burkay adds that sophisticated users exist but broad adoption is still developing.
- •Molly discusses ads as a major emerging application and current limitations in generation length
- •Görkem says UGC-style ads are a strong fit because perfection isn’t required and iteration is cheap
- •He suggests ads can be assembled from short cuts rather than one continuous long generation
- •He highlights the opportunity to optimize creative based on real performance data
- •Burkay says some companies already do this, but technical complexity limits widespread use today
- 13:20 – 18:02
fal’s origin story: from compute thesis to running generative media models fast
Burkay explains fal started as a compute-focused company in a pre-generative-AI era, aiming to bring Python-based workloads to the cloud. After breakthroughs like ChatGPT and image models, they chose to pursue image-model acceleration rather than compete in a token race, then named and framed the category as “generative media.”
- •fal began in 2021 with a compute-centered thesis inspired by prior infrastructure waves
- •Early focus was enabling Python workloads in the cloud for data science/ML teams
- •They saw market breakthroughs (ChatGPT and image models) and made a strategic pivot
- •They chose image models as a more interesting direction versus low-margin token selling
- •They coined and adopted the term “generative media” to explain the company’s focus
- 18:02 – 24:28
Staying ahead of research by hiring fanatics and building a leadership-driven learning loop
Molly asks how they keep up with fast-moving research and model releases. Görkem emphasizes relying on a strong leadership team, while Burkay describes recruiting people who “live and breathe” the space—drawing inspiration from Coinbase’s early hiring approach to find true believers.
- •Görkem says early on they were deeply involved, but now lean on a trusted leadership team
- •He says he learns more from internal leaders than from the internet today
- •Burkay describes Coinbase’s early “crypto quiz” as a filter for deep interest and knowledge
- •They prioritize hiring people who would be doing the work anyway because they’re obsessed
- •They note generative media was initially overlooked because it rose alongside LLMs
- 24:28 – 27:49
Co-founder relationship and early team formation from trusted networks
The founders explain how they met and why their long friendship mattered when starting the company. They describe pulling early hires from prior workplaces and argue that starting later in one’s career can help because you’ve built trust with strong people over time.
- •They met in San Francisco after Görkem moved there and became longtime friends
- •They tested working chemistry during a shared Palm Springs house rental in COVID
- •They started the company about a year later and have been building for years since
- •They hired from both Coinbase and Amazon
- •Görkem says starting in their mid-30s helped because they could recruit trusted former coworkers
- 27:49 – 32:28
Flat org design, "Super ICs," and using agents to scale without pure management layers
Molly asks about scaling vs staying flat, prompting a discussion of fal’s organizational philosophy. Burkay defines “Super ICs” as hands-on leaders rather than pure managers, while Görkem argues improving coding models make it possible for small teams to do more with fewer layers.
- •Burkay says fal is flatter than typical startups, with limited reporting depth
- •They prefer “Super ICs”: leaders who both execute and lead teams
- •They aim to avoid pure management roles and let leaders emerge naturally
- •They acknowledge communication and alignment bottlenecks as scaling challenges
- •Görkem ties the approach to coding agents enabling higher leverage per leader
- 32:28 – 35:58
Hiring traits: agency, hunger, humility—and realistic expectations about workload
The conversation turns to what they screen for in candidates and what makes the best fits in a fast-moving market. Görkem emphasizes “agency” as the ability for engineers to make decisions and take ownership, while Burkay adds “hunger” and “low ego,” and describes using reverse-selling to set expectations.
- •Görkem says agency is critical because engineers often must make business decisions
- •He notes agency is hard to test without having worked together before
- •Burkay highlights hunger as essential for pushing to win in a fast market
- •He describes reverse-selling by being candid about time crunch periods
- •Burkay says they value humility and being easy to work with, even among highly talented people
- 35:58 – 39:30
Compute constraints and the infrastructure role: planning model launches and ecosystem supply/demand
Molly asks about challenges and patterns in model quality, leading to a deeper discussion of staying at the frontier. Görkem describes fal’s dual role serving model providers and the application layer, arguing that compute is scarce and requires proactive planning and investment to support launches and growth.
- •Burkay says staying at the forefront takes commitment and sustained effort
- •Görkem describes the industry as highly dynamic, with rapid shifts in expectations
- •He explains fal partners closely with model providers for inference and launches
- •Compute scarcity forces providers to plan ahead and often partner rather than self-procure
- •fal must anticipate growth and invest in compute so generative media isn’t crowded out
- 39:30 – 45:12
Why model ecosystems are fragmented, why names get weird, and what “fal” means
They explain why generative media has many modalities and frequent leapfrogging among many labs. This leads into a playful detour about funny model names and code names, followed by the origin story of the name “FAL,” including its meaning as “Features and Labels” and its Turkish cultural reference to fortune-telling from coffee grounds.
- •Burkay lists multiple generative media modalities beyond video (image, editing, 3D, speech, etc.)
- •They describe frequent leapfrogging and many competing players
- •Görkem says abundant video data and broader problem space drive continued progress
- •They discuss model naming, including code names used during anonymous releases
- •They explain FAL as “Features and Labels” and as a Turkish coffee fortune-telling reference
- 45:12 – 48:53
Open vs closed source: marketing, customization needs, and trust/security for studios
Molly asks whether the open-source vs closed-source debate will continue. Görkem argues it will persist because open sourcing is both a marketing tactic and enables studio-specific customization; both founders note companies want to protect IP, and Görkem emphasizes trust and security—especially for unreleased content—as a key enterprise differentiator.
- •Görkem says open sourcing is a powerful marketing move for those falling behind
- •He says studios want fine-tuning, LoRAs, and style control—easier with open source
- •He notes fal benefits either way by running inference for both open and closed models
- •Burkay says IP protection concerns will always keep the debate alive
- •Görkem stresses trust/security matters because creatives may input unreleased content
- 48:53 – 1:03:36
Hollywood reality check, near-term content milestones, and hot takes on margins and industry structure
They push back on the narrative that AI has already transformed Hollywood, with Görkem saying it has had “zero impact” so far despite growing business interest, partly due to legal/regulatory friction. They close with what they’re excited about (evergreen AI content and world models), who inspires them, and hot takes—especially that application-layer margins are near zero and compute-path innovation is where margin exists.
- •Görkem says AI has had zero impact on Hollywood so far, despite fearmongering
- •He attributes slow adoption partly to legal approval requirements, now easing تدريجياً
- •Görkem looks forward to the first "timeless" AI content people still enjoy a year later
- •Burkay is excited about world models and persistent interactive worlds, but notes compute challenges
- •Hot takes: Burkay speculates an OpenAI–Meta merger could make sense; Görkem says application-layer margins are near zero and compute path is where margin is created
