Anton Osika on Lovable’s Launch, Early Retention and Building AI Products in Europe

Lovable co-founder and CEO Anton Osika says the AI app-building company’s early growth came from fast iteration, a strong team and a product people wanted to use. In this March 2025 interview, Harry Stebbings asks about launch, fundraising, retention and the future of software. Osika’s descriptions and forecasts reflect his position at Lovable.

At his previous company Depict, Osika says the team moved quickly and hired promising junior talent. But it said yes to too many opportunities. Once the organization grew, the work of maintaining everything made it harder to focus on one thing the company could do exceptionally well.

For product design he favors saying no to as many things as possible and making the chosen features purposeful. He considers talent and the culture of daily collaboration the two most important ingredients for most companies.

Osika often favors potential over experience: ambitious junior people can be open to new ways of working and may later become founders themselves. He still wants the relevant engineering skills, and says some roles need experienced people who can coach others and establish standards.

He says he was often frustrated when others did not grasp ideas as quickly as he thought they should, but later learned that he could also be naive. As a teenager he earned money helping neighbors and family friends with computer problems.

After ChatGPT emerged, Osika experimented with the idea of placing a language model in a loop to perform agent-like tasks. He chose software engineering as an application and built an early open-source GPT Engineer prototype that could generate a working snake game.

The first version took one intensive weekend and some later polishing, he recalls. His general advice to first-time founders is to solve a real user problem and make one person love the first version. In this case, he posted a video and the open-source project attracted a large community.

Initially he was not sure the project would become a business. Its response persuaded him to find a replacement for his CTO role at Depict and recruit a co-founder. Together they began building the product that became Lovable.

Before the November 2024 Lovable launch, the team released waitlisted previews under the GPT Engineer app name. Those versions helped gather feedback and build interest, but Osika says the value was not yet clear to enough people. The team iterated over the following year.

A waitlist lets a team control the number of users it onboards and choose whom to interview. Osika suggests identifying the users most likely to benefit, qualifying them and learning what problem they actually want solved.

He distinguishes observing a person use the product from interviewing them about the business problem behind their interest. Watching reveals usability problems; asking about their work reveals whether they want, for example, to win more customers or deliver an initial version faster.

For a personal website or first product, he believes one person using AI may be faster than coordinating a team. Once software has users, however, an AI change can damage an existing codebase; an engineer may be needed to maintain quality.

Osika says time to the first moment of value is important and admits Lovable could improve it. He speculates that better onboarding could double conversion. The product starts with an interactive prompt box, which he believes gives users an immediate reward; more complex uses require further learning.

He thinks prompting will remain useful because people can express many intentions in writing and teams can iterate on the interface easily. He also expects interfaces to become more advanced over time; that is a prediction, not a settled design standard.

The founders declined Y Combinator because they thought the dilution might outweigh acceleration and the program could distract them from building. They instead took funding and focused on talent.

Before the first waitlist, Osika says Lovable raised an initial $3 million and then expanded the pre-seed to almost $8 million. He wanted investors he liked working with in both good and difficult times, and accepted a preemptive offer.

A larger early round can provide time to build without frequent fundraising, he says, while founders who enjoy investor conversations may prefer smaller iterative rounds. He weighs dilution against the size of the eventual opportunity and the fact that this is his life’s work.

Lovable launched on November 21, 2024, with some paying users already on an earlier version. Osika says the launch itself could have attracted more press, but users noticed the product and the team kept improving it quickly.

Growth accelerated after launch: he says annual recurring revenue increased by about $1 million per week in December, then by roughly $2 million per week at the time of the conversation. These are company-reported changes in annualized recurring revenue, not weekly cash revenue or independently verified results.

Rapid growth also created scaling problems. The team chose to rewrite much of the system while fixing product issues; Osika says the work took more than eight weeks and was not fully complete, but the team was subsequently shipping faster.

Complex products and too many requirements slow development, he says. He regrets spending time on community features inside Lovable: with growth already strong, he no longer considered them necessary to drive it.

“Build it and they will come” can work when a founder has a strong, tested vision and enough personal and financial runway, Osika says. For most teams he considers it risky and recommends building while actively trying to reach users.

Reflecting on Depict, he says a major first-time management mistake was adding management layers and hiring executives when the company reached roughly 40 people. One hire did not work out and slowed the team. He now favors empowering capable generalists, though he accepts that executives can make sense at a later stage.

Rapid growth can change or dilute culture, he says. At Lovable he wants people to model care for the product, users and one another, and he is cautious about adding headcount so quickly that a sense of ownership disappears.

Although the company was generating revenue, it raised a small round after Osika developed a strong impression of Fredrik at Creandum. He valued help finding talented people and a sounding board, and believed the partnership could accelerate the business. His description of the investor reflects an existing funding relationship.

Osika does not believe a well-funded competitor automatically forces a company to raise more money. He sees execution as the decisive factor and says Lovable could improve its decision speed, communication and focus by doing fewer things at once.

He intends to build in Europe, arguing that the region has considerable available engineering talent. He also sees Silicon Valley as having a default culture more oriented toward ambition and startup commitment, while European norms often favor balance. These are his impressions, not universal facts.

He calls building from Europe “hard mode” in some respects but likes the challenge. He is optimistic about European founders and believes companies can use European talent while selling into the US. His assessment of relative talent costs and opportunity is a business view.

Stebbings asks whether AI revenue is merely short-lived experimentation. Osika says Lovable’s month-one retention among paying customers was about 85% and claims it exceeded ChatGPT’s corresponding paying-customer measure. The interview does not specify the cohort, measurement method or comparable external data, so the cross-product claim cannot be independently assessed.

He acknowledges that some people pay to try the product and expect to leave after learning about it. The company was too young in this conversation to offer six-month retention evidence. An early month-one figure does not establish long-term retention.

To improve retention, he wants users to experience more moments of value. Beyond seeing a first app generated, users need to learn how to clarify prompts when the AI gets stuck and when an engineer can help make small changes to a codebase.

Lovable’s north-star metric, he says, is users who publish something and attract users of their own. He reports almost 40,000 paying users as a proxy, not as the same measure. The figure is a company-reported snapshot from the interview.

Asked whether Lovable is merely a wrapper over other models, Osika says a demo is easy but reliably producing the requested result requires many details, chains of model calls and other algorithms. He says the system can be improved for years without reaching perfection.

Lovable used models from OpenAI and Google Gemini, with Anthropic’s Claude as the main code-writing model at the time. Asked to compare model companies as investments, Osika favors the team behind Grok, praises Anthropic’s culture and says he would short OpenAI. These are hypothetical personal views, not investment advice.

Stebbings counters that ChatGPT has strong brand recognition and a consumer product lead. Osika agrees those are strengths but thinks the market remains early and cites his view of enterprise competition. The interview does not independently establish the companies’ relative revenues.

He sees the distribution of large incumbents as a greater concern than their speed of product development. He believes a growing market leaves room for a startup with a strong product, while recognizing that a competitor exceptionally good at marketing could threaten Lovable.

At the time, Lovable was holding off on enterprise sales because Osika wanted to do that well rather than enter prematurely. Its goal was to become the best place for builders to create products and attract a million talented builders, which he saw as a possible route into enterprise later.

Stebbings admires Shopify’s ability to tell the stories of entrepreneurs and show how the product affects their businesses. Osika emphasizes Shopify’s rapid iteration and the way it assembled the capabilities needed for e-commerce.

One product idea on Lovable’s roadmap is to give founders more of the infrastructure and playbook associated with starting a company, potentially including incorporation and payment setup. Osika describes this as a future direction, not functionality established in the interview.

Further fundraising would be worthwhile, he says, if the company knew how to deploy the capital or found a partner and terms it would not regret. At that point he regarded most investor approaches as a distraction and thought one or two exceptional technical product hires could be more consequential.

In the quick-fire round, Osika expresses his belief that models are already very intelligent but lack comparable memory and context. He expects building useful memory systems to take years; these are his judgments about a developing technology.

Asked for a public stock to hold for ten years, he mentions Tesla as a talent-oriented choice. He values founder judgment in identifying potential in people, and says his own weakness as CEO is multitasking. His stock choice is a personal view, not investment advice.

A possible reason Lovable could fail, he says, is losing the momentum and excitement that currently fuel it. He has also changed his mind about dependence on a single foundation-model provider and predicts greater commoditization across models.

He calls Depict not working out a favorite failure because it made room for subsequent projects. Asked what worries him globally, he criticizes narrow-minded leadership and wants leaders who hold ideals while listening to people who disagree.

He suggests that per-seat software businesses could face pressure if AI agents replace some of their customers’ employees. For defensible enterprise products he considers value-based pricing an option, while admitting it is difficult to design; otherwise he suggests using pricing that already works.

Looking ahead, Osika hopes AI will help people understand each other and find mutually beneficial outcomes. He is excited by the possibility of leadership supported by more capable AI, a hope rather than an established outcome.