The Twenty Minute VCAnton Osika, Co-Founder and CEO @ Lovable: Hitting 85% Day 30 Retention - Better than ChatGPT
CHAPTERS
- 0:00 – 0:51
Lovable’s retention claim and rapid growth in context
The conversation opens with Anton’s headline metrics: Lovable’s month-one retention and the pace of ARR growth post-launch. Harry frames Lovable as Europe’s fastest-growing company and tees up themes of talent, culture, and execution speed.
- •Month-one retention cited as better than ChatGPT
- •Early mention of explosive ARR growth trajectory
- •Positioning Lovable as a standout European hypergrowth story
- •Talent and culture introduced as core drivers
- 0:51 – 2:59
Lessons from Depict: speed vs focus, and the cost of saying yes
Anton reflects on scaling quickly at Depict and what worked—scrappy execution and hiring high-potential junior talent. He contrasts early-stage experimentation with the need for ruthless focus as teams grow, and notes how macro conditions exposed the downsides of overextending.
- •Early: say yes, move fast, explore opportunities
- •Later: focus becomes essential because everything requires maintenance
- •Depict mistake: too many initiatives vs one thing done 10x better
- •Macro downturn revealed fragility in the scaling plan
- •“Depict mafia” highlights talent density and alumni impact
- 2:59 – 4:37
Hiring philosophy: why talent can beat experience (and when it can’t)
The discussion dives into Anton’s belief that experience can sometimes be a liability, especially in fast-moving environments. He explains why ambitious, adaptable junior hires can outperform, while acknowledging specific domains where experienced leaders are necessary to set quality bars.
- •Experience can reduce open-mindedness and adaptability
- •Junior, high-potential hires are easier to attract and shape
- •Best young talent often becomes founders—hire them early
- •Some roles require experienced coaches to define “great”
- •Culture and collaboration speed are treated as competitive advantages
- 4:37 – 6:31
Founder origins: early hustle, gaming, and “future-truth” intuition
Harry explores Anton’s early signs of founder behavior—making money young and gaming—then Anton describes his first side hustle fixing computers. Anton also shares his self-perceived superpower: strong intuition about what’s coming next, even before he knew it would translate into company-building.
- •First money: troubleshooting computers for neighbors and friends
- •Gaming as a recurring founder archetype
- •Anton’s “track record” for predicting future trends
- •Learning curve: balancing confidence with avoiding naivety
- 6:31 – 9:00
From GPT Engineer side project to Lovable: the agentic coding insight
Anton recounts how, after ChatGPT’s release, he recognized the wave from scaling models and began thinking in terms of agents (LLMs in loops). He built the first GPT Engineer prototype quickly, demonstrating real code generation (e.g., a working Snake game) to prove feasibility.
- •Core insight: agentic behavior via LLMs in a loop
- •Applied to software engineering as the obvious use case
- •Built v1 in a weekend plus a couple follow-up weekends
- •Early demo value: prompt → runnable application
- •Initial motivation: prove the point, not start a business
- 9:00 – 10:22
Open source momentum, leaving Depict, and recruiting the co-founder
What began as an open-source project attracted a community, which created momentum and clarity that Anton should move on from Depict. He describes finding a replacement CTO and then directly recruiting a highly efficient entrepreneur-engineer as co-founder to start building the Lovable product.
- •GPT Engineer started open source and community-driven
- •Wake-up call: time to transition out of CTO role at Depict
- •Recruiting approach: direct, personal, high-conviction outreach
- •Co-founder fit: “zero fluff,” execution-oriented
- •From community project to company-building path
- 10:22 – 12:57
Pre-launch waitlist strategy and user interviews: controlling learning loops
Anton explains how Lovable used waitlisted preview releases (GPT Engineer app) to manage onboarding volume and learn from the right users. He breaks down two distinct interview modes—observing product usage for UX clarity, and probing underlying business pains to understand true value drivers.
- •Waitlists help throttle demand and focus on high-signal users
- •Qualify who to talk to based on hypotheses about value
- •Two interview types: usability observation vs needs/pain discovery
- •Aim: identify what users truly want to achieve (not just features)
- •Early challenge: users needed help reaching clear “aha” moments
- 12:57 – 15:51
Product UX and the “time-to-aha”: prompt box UI, but more than one aha
They discuss how AI changes team formation—solo builders can ship faster, while mature codebases still need engineering rigor. Anton emphasizes reducing time-to-aha and explains Lovable’s choice to lead with an interactive prompt box, while acknowledging there are multiple aha moments (especially when users get stuck).
- •AI enables solo creation of first versions; teams can slow early progress
- •Risk: AI can degrade code quality in existing systems without oversight
- •Time-to-aha is critical; Lovable can improve onboarding and conversion
- •Landing experience: immediate interactive prompt box
- •Multiple “aha” moments: recovering from stuck states, better prompting, hybrid AI+engineer workflows
- 15:51 – 18:47
YC decision and pre-seed: raising early, choosing investors you trust
Anton explains why they rejected Y Combinator—viewing it as dilution and potential distraction—and instead raised funding to stay focused. He outlines a founder-centric investor approach: work with people you genuinely like and trust, accept preemptive offers quickly, and raise enough cash to weather market changes.
- •YC tradeoff framed as acceleration vs dilution/distraction
- •Seed/pre-seed came before even the first waitlist launch
- •Investor selection: prioritize character and long-term partnership
- •Round size: started at $3M, extended up to nearly $8M
- •Debate: raise big for focus vs smaller iterative fundraising if you enjoy it
- 18:47 – 21:34
Public launch to hypergrowth: ARR acceleration, scaling pain, and rewriting fast
Anton recounts going live on Nov 21 and describes growth ramping after launch as users discovered product quality. He shares eye-watering numbers (from $1M to $2M ARR per week) alongside the operational reality: severe scaling issues forced an aggressive rewrite, after which shipping speed improved.
- •Launch date: Nov 21; early press could have been bigger
- •Growth ramped post-launch via product pull and rapid iteration
- •Reached ~$1M ARR/week, then ~$2M ARR/week
- •Scaling issues prompted an ~8-week rewrite (still ongoing)
- •Lesson: technical foundations and speed of fixes matter during hypergrowth
- 21:34 – 26:55
Execution and culture at scale: simplicity, avoiding exec layers, and decision loops
They unpack what slows product development—complexity and too many requirements—and Anton admits some effort (community features) was unnecessary given growth. He also reflects on a major Depict mistake: adding management layers and exec hires too early, which can dilute ownership and slow smart generalists.
- •Complexity and requirements are the main enemies of speed
- •With hindsight: community features were low ROI given strong demand
- •“Build it and they will come” only works with strong conviction + runway
- •Biggest past mistake: trying to “scale up” culture at ~40 people
- •Advice: be cautious adding exec layers over high-agency generalists; improve decision/communication loops and do fewer things at once
- 26:55 – 29:27
Series A rationale and competitive posture: money doesn’t beat execution
Anton explains raising a (small) Series A primarily because of a high-conviction partner (Creandum/Fredrik) who could help with hiring and strategic feedback. They debate whether competitor war chests force fundraising; Anton argues execution is the only real threat, not spending, and highlights focus as the key execution improvement.
- •Series A driven by investor quality and leverage (hiring, sounding board)
- •View: you can bootstrap; fundraising isn’t mandatory because others raise
- •Not afraid of being outspent on marketing/talent—fear is being out-executed
- •Execution upgrade areas: faster decisions, clearer communication, less parallel work
- •Needle mover: a small number of “perfect” technical product hires
- 29:27 – 32:04
Building from Europe: talent advantages, cultural tradeoffs, and the underdog thesis
The conversation turns to geography: Anton defends staying in Europe as a deliberate choice, emphasizing raw talent availability and the challenge/opportunity of building “on hard mode.” They discuss cultural differences (ambition vs balance), and the strategy of selling globally—especially into the US—from a European cost base.
- •Europe’s edge: abundant raw engineering talent
- •US edge: startup-optimized ambition and risk culture
- •“Hard mode” motivation: prove category-defining companies can be built in Europe
- •Optimism: underdog mentality can be a competitive advantage
- •Global market reality: build in Europe, sell into the US
- 32:04 – 35:42
Revenue durability, retention metrics, and the real North Star for Lovable
Harry challenges whether Lovable’s growth is “AI sugar revenue” that will churn; Anton counters with 85% month-one retention on paying customers and a rising trend. He clarifies that improving retention is about teaching users advanced success patterns when they get stuck, and shares the North Star: users getting real end-users on what they built (proxied by paying users).
- •Critique addressed: sustainability vs hype-driven churn
- •Metric: ~85% month-one retention among paying customers; trending upward
- •Churn segment exists: users who pay just to experiment briefly
- •Retention lever: educate users through stuck points and complex workflows
- •North Star: successful hosted projects with real users; scale proxy: ~40,000 paying users
- 35:42 – 49:39
Moats, model dependence, and the foundation model “buy/short” debate (plus quick-fire)
Anton explains why Lovable isn’t just a wrapper: reliability requires a complex chain of model calls and algorithms refined over time. He discusses multi-model usage (OpenAI, Gemini, Claude), then shifts into broader predictions—foundation model commoditization, model memory timelines, pricing impacts—and a quick-fire set including public-market takes and long-term optimism about AI improving human coordination.
- •Moat claim: hard part is near-100% correctness, not demos; requires complex orchestration
- •Model stack: OpenAI + Gemini + Claude (workhorse for code)
- •Investment take: buy Grok (talent/ruthless execution), short OpenAI (focus/product direction concerns)
- •Belief: foundation models commoditize; no single permanent winner
- •Quick-fire themes: AI smarter-than-human in some ways but lacking memory; per-seat SaaS risk; leadership concerns; decade optimism—AI enabling better win-win outcomes
