Anton Osika, Co-Founder and CEO @ Lovable: Hitting 85% Day 30 Retention - Better than ChatGPT

Summary

• Lovable is adding $2M in ARR every week, up from $1M/week in December, after launching on November 21, 2024 — and Anton Osika admits the launch itself was underwhelming (“we could have gotten 10 times more press”). Growth accelerated even while the team spent 8+ weeks rewriting the entire codebase under load.
• The answer to the “AI sugar revenue” critique is the sharpest data point in the episode: 85% month-one retention on paying customers, better than ChatGPT’s, alongside almost 40,000 paying users — with Osika conceding some users knowingly “flip up their credit card” just to learn, and Harry noting that it’s “too early to have month six.”
• Osika’s capital stance is genuinely contrarian: he rejected YC (“at best a lot of dilution and some acceleration, at worst a distraction”), later raised a small round he says he could have raised later, and says well-funded US competitors don’t force you to raise — “the only thing that matters is execution… you can bootstrap most things.” He’s not afraid of being outspent on talent, customers, or marketing.
• His asset-allocation quickfire is tradeable-adjacent: buy likely Grok at $50B, short OpenAI at $300B — Elon is “very good at talent,” while OpenAI “lost all their best talent to Anthropic” and hasn’t shown clear product direction — even though Anthropic (whose Claude is Lovable’s “main workhorse” for writing code) is his favorite. Biggest public-market short: per-seat SaaS whose ICP gets replaced by AI, because “the number of seats goes down.”
• A change of mind worth flagging: you don’t need to attach to one foundation model provider — “they’re all going to be amazing, there’s not going to be one winner” — and current specializations (Claude best for code) will equalize as models fully commoditize. Lovable already runs across OpenAI, Gemini, and Claude.
• On Europe: “there’s more raw available talent in Europe” at arbitrage pricing selling into the US, even if US culture better defaults to thinking big versus Sweden’s law of Jante. Building a category-definer from Europe is “playing on hard mode — and I get excited about playing on hard mode.”
• The self-critique is the growth lever: Lovable is “very bad at making the time to aha moment super short” and could double conversion rates by fixing it — the team has focused on “making the core AI parts better better better,” not onboarding. Premortem: the company dies if it loses “momentum and excitement — that’s what fuels us.”

Deep dive

1. $2M new ARR a week, 85% month-one retention — the numbers behind “Europe’s fastest scaling company”

• The trajectory as Osika tells it: launch on November 21, 2024 (“that was only four months ago”), growth starts ramping after launch, reaches $1M ARR added per week at some point in December, “and that just keeps accelerating” to $2M per week now — with almost 40,000 paying users. Harry contrasts it with going “one to four in a year” — a flat non-answer: “a lot of things the last few weeks have just blunted me… I’m just focused on all the things we have to fix.”
• Against the “sugar revenue” critique Harry raises directly, Osika’s rebuttal: “we have month one retention that’s better than ChatGPT’s” — about 85% on paying customers, and going up — while conceding a cohort of users “flip up their credit card because they want to try… they almost know” they’ll churn. Harry notes that they are “too early to have month six.”
• The scaling challenge during the surge: explosive growth hit scaling issues, and the team chose to rewrite everything mid-surge — “it took a bit more than eight weeks… and now we’re shipping faster.”
• North star metric: not revenue but users who get something hosted with real users on it — people who “built their entire SaaS companies and made money by just prompting our AI.”

2. The origin: “put the large language model in a for loop”

• GPT Engineer was conceived on Osika’s engagement trip, the spring after ChatGPT launched: he didn’t think anyone was talking about agents yet, but on the plane he sketched the core insight — “you basically put the large language model in a for loop and then you can have it do a lot of agentic things” — and felt “no one was really sufficiently imaginative to what I was thinking about.”
• Execution as told: “I drank a lot of coffee and then I just crammed away” — V1 in one main weekend plus polish over two more. The demo — type “create a snake game,” get a running snake game — went out as a Twitter video and drew “dozens of academic references and millions of people using it,” and he did not follow his own advice to first make one person love the V1.
• The co-founder recruitment is a specimen of his talent thesis: he identified “the most super efficient, zero fluff” engineer who had already sold a company, biked to his apartment and said “let’s take a walk and plan the future.”

3. Depict lessons: say no, hire junior, fire the executive playbook

• The Depict lesson that shaped Lovable: Depict “said yes to too many things” and didn’t take “this one thing that we could do 10 times better than anyone else.” He endorses Paul Buchheit’s three-great-features rule — “say no to as many things as possible and make it more of an Apple feeling.”
• On talent, “experience can be a negative thing”: junior high-potential hires are ambitious, have a lot to prove, open-minded, and uncommitted — “the best people you can hire at a young age would go on and become founders and then you can’t hire them anymore.” The “Depict Mafia is absolutely real.”
• His self-declared biggest mistake as a first-time manager: hiring executives at 40 people because senior advisors said to. “Don’t believe the [bs] — you can scale way longer without execs.” His model instead: super-smart generalists, empowered; adding executives on top of them is “high risk and the reward is questionable.” It’s also why he’s “so scared of adding too many heads” now — ownership of culture “gets diluted” with headcount.

4. The anti-fundraising fundraise: rejecting YC, ignoring war chests

• YC was rejected on a clean expected-value read: “at best a lot of dilution and some acceleration, and at worst a distraction.” Instead: a preemptive ~$3M seed grown to “almost $8 million” pre-launch, from investors he simply liked — kept “very, very brief.”
• The later round could have waited — “I could raise it later” — but Creandum’s Frederick, who “helped grow Spotify from nothing,” was worth adding as a partner and sounding board, so he “decided to raise a small round.”
• Harry’s spiciest structural question — must you raise when competitors are loaded? — gets a categorical no: “you can bootstrap most things, so you never have to raise.” Outspent on talent, customers, marketing? “I’m not afraid of any of those. The only thing that matters is execution.” New money now is “currently a distraction”; the real needle mover would be “one or two more of the perfect technical product hires.”
• On dilution sensitivity, he holds two ideas at once: the wise-person advice that “dilution doesn’t matter so much, it’s all about the size of the pie” — and “this is my life’s work,” so minimize it anyway.

5. Product self-critique: the aha-moment gap could double conversion

• His most honest admission: “we are very bad at making the time to aha moment super short — I think we could double our conversion rates” by fixing it. What works: landing users straight into a prompt box, because you should “give the user something interactive with instant reward.” What’s not been a focus: onboarding — “we were just making the core AI parts better better better.”
• When Harry pushes back that the aha moment is obvious (click a prompt, watch code appear), Osika disagrees: the important aha moments are for when “you’re getting stuck and the AI doesn’t understand you” — how to prompt, how to explain what’s broken, how to onboard an engineer for small codebase changes.
• The wrapper critique gets a mechanism, not a dismissal: “it’s super easy to make a cool demo with just a wrapper. The hard part is to get close to 100%” — a chain of LLM API calls and algorithms “you can continue to optimize for years without reaching perfection.” Confirmed wasted effort: in-product community features, “pretty much a waste” once growth wasn’t the constraint.
• On chat as the default UI, a hedged yes after a long pause: prompting “remains” but gets more advanced — “you’re building the interface for creating software and no one knows what that interface is going to look like.”

6. Europe on hard mode

• The case for staying: “the most important thing is talent and culture, and there’s more raw available talent in Europe” — plus “incredible superpowers in using the arbitrage pricing of incredible engineers in Europe and selling into the US.”
• The concession: US culture defaults to “thinking big and being super ambitious,” while Sweden has the law of Jante and a preference for balanced living. His resolution: “it’s a bit of playing on hard mode here from Europe, and I get excited about playing on hard mode” — and the underdog mentality among European founders now is “usually a winning concept.”

7. Market calls: buy likely Grok, short OpenAI, short per-seat SaaS

• Given Anthropic at $60B, OpenAI at $300B, and likely Grok at $50B: buy likely Grok (“I care about the best talent here and Elon is very good at talent… they’re also very ruthless in finding business opportunities”), short OpenAI — good in “the scrappy phase” but no proven product direction, and “OpenAI lost all their best talent to Anthropic,” which has “almost caught up to OpenAI” in enterprise revenue “from being absolutely dominated.” This despite Anthropic being his favorite and Claude the “main workhorse” writing Lovable’s code.
• Harry’s pushback — worth keeping: the next wave is won on brand and consumer product, “everyone’s mother knows ChatGPT,” and OpenAI is “way ahead of anyone else” on consumer. Osika’s only rebuttal: “we’re still early in the days of AI.”
• His 12-month change of mind: “you don’t need to be attached to one foundation model provider — they’re all going to be amazing, there’s not going to be one winner.” Foundation models fully commoditize, and today’s specializations (Claude for code) will equalize. Mega-corp distribution worries him more than their product: “they’re not going to have the best product for many years,” but marketing and distribution advantages are real.
• Public-market picks: ten-year hold is Tesla as “some talent play… interestingly positioned outside of software”; biggest short is a per-seat SaaS company whose ICP gets replaced by AI — “the number of seats goes down.” Enterprise for Lovable is deliberately deferred: the goal is “a million of the most talented builders on Lovable,” and building YC-in-a-box — Delaware C-corp, Stripe setup, marketing playbook — is “on the roadmap.”