Lovable CEO Anton Osika on AI App Building, Economics and Security

Harry Stebbings interviews Lovable co-founder and CEO Anton Osika about AI models, application building, security and company culture. The episode description identifies 20VC as a Lovable investor, and the host says he invested. Osika speaks for his company, so his competitive comparisons and forecasts carry direct commercial interests.

Osika says AI application companies compete to build strong teams and earn users’ trust, with capital as an aid rather than Lovable’s current constraint. He distinguishes the skills needed to build an application from those needed to train frontier models. That is his management view, not a ranking of individual engineers.

Asked how a startup can compete with large talent packages, he says the harder problem is identifying people who will thrive together. He looks for learning speed, dynamic conversation and evidence of past work, and says compensation must still be competitive. His assessment process is a preference, not a validated hiring test.

Osika says he still operates in a scrappy founder style but wants more structure to prioritize the many demands on him. Stebbings questions whether additional management layers slow companies down. Osika says close collaborators can provide organization while he remains involved in the work.

He uses Apple as an example of brand built through attention to details and says Lovable wants users to trust repeated product interactions. His ambition is to build a platform where users have created enough value that they want to stay; the interview does not provide customer retention data to prove this.

Stebbings asks about defensibility. Osika advises early AI startups to execute quickly and develop defensibility over time, arguing that new competitors arrive constantly. His proposed expansion from technical co-founder to broader business platform is a future strategy, not an established capability.

Asked about unit economics, Osika says the majority of paid usage spending passes to model providers, but declines to provide exact figures. He hopes subscriptions and services used after initial building will improve the economics. This is a company strategy and does not establish current gross margin or future profitability.

The host asks whether smaller models could handle simple tasks more cheaply. Osika says Lovable prioritizes adapting to rapidly changing capabilities over optimizing every model cost immediately. He expects a mix of fast and deeper reasoning systems, depending on the task; the cost and quality trade-offs remain workload-specific.

Osika says that, a few months earlier, more than $10 million in annual recurring revenue was flowing through AI used by applications built on Lovable. He describes simplifying users’ connections to model providers and possibly earning a margin. This is not clearly stated as Lovable’s own ARR or revenue and should not be added to its reported company figures.

He weighs advice to optimize customer acquisition payback and margins against his current priority of building user loyalty and brand. Stebbings favors the latter early on. The right balance depends on actual acquisition costs, retention and capital; neither provides a financial reconciliation.

Osika wants software built with AI, payments and other capabilities integrated more seamlessly. He expects model providers to enter Lovable’s market and sees OpenAI as a particularly serious potential competitor. His claims about the product’s future and competing teams are interested forecasts.

Discussing GPT-5, he says Lovable evaluated response times, quantitative tests and user feedback. He describes it as useful for difficult problems but sometimes too ambitious for the product’s users, with trade-offs from combining capabilities. His observations are specific to Lovable’s tests and are not a general ranking.

He says Lovable uses a chain of different models, including Anthropic for much code writing and GPT-5 for some difficult debugging. He hopes future systems will use more context about each user. The product design and model comparisons are the company’s account at the time.

Stebbings says Lovable reached $100 million in annual recurring revenue in seven months; the episode description reports $120 million ARR. These are company-associated, time-specific annualized recurring-revenue figures, not realized annual revenue, net income or cash flow. The interview does not provide audited statements or a complete reconciliation.

Asked about customer mix, Osika estimates roughly 80% of revenue comes from people building substantial applications, with enterprise and hobbyist use roughly 10% each. He agrees these are approximate categories and expects the mix to change as enterprise use grows. No segmentation methodology is disclosed.

Stebbings asks whether helping founders build businesses offers enough ways to capture value compared with a huge consumer market. Osika emphasizes enabling people without coding skills or capital to create businesses, and expects a broader audience to follow. The opportunity and monetization are his forecasts.

The host explicitly thanks Osika for accepting his investment and argues that the addressable market could expand far beyond website builders. Osika agrees there may be many future companies built with AI. The host’s investment and the CEO’s position are relevant to their optimism; the market size is not measured in the conversation.

Osika imagines enterprise teams using working prototypes to bring ideas to products faster. He argues Lovable has compressed early steps of the product lifecycle and wants to address later stages too. The interview offers examples and aspirations, not controlled evidence of productivity across companies.

Stebbings asks whether Figma can move from design into prototyping and threaten Lovable. Osika predicts more high-level design direction and AI implementation, while acknowledging that detailed design tools may remain useful. He does not know how the market will divide; his company competes in the area.

He expects prompting to continue but become more personalized as systems understand users’ preferences and context. That is a product prediction, not evidence that models can reliably infer a user’s intent or safely handle all personal context.

The speakers discuss public criticism between competitors after security issues in generated apps. Osika describes reacting to a competitor’s disclosure; Stebbings challenges all the companies on security. Neither side supplies a comprehensive independent security comparison.

Osika acknowledges Lovable is not yet more secure than human development generally. He says the product prompts users through reviews and claims it can reduce vulnerability risk compared with an average developer working alone. This is a company claim without disclosed comparable testing; a clean automated review does not establish zero vulnerabilities.

He says his long-term aim is a world in which people need not write code, but accepts that users today may combine Lovable with human engineering and other tools. His expectation that a single platform will eventually be the obvious choice is an interested prediction.

Asked whether AI helps junior or senior engineers more, Osika says it can help both in different contexts: it may bridge a novice’s missing skill or accelerate an expert who understands a complex system. These are illustrative judgments, not measured productivity multipliers.

He expects engineers to become more product-oriented and generalist as AI handles more detailed tasks. Whether headcount grows or shrinks depends on demand and company choices; the interview does not establish future employment effects.

On studying computer science, Osika questions whether university is the best place to learn and emphasizes practical work, while acknowledging its social and educational breadth. Stebbings shares his own skepticism. Their personal opinions do not determine the value of a degree for other people or professions.

Stebbings asks whether established enterprises will lose ground because of data permissions and security constraints. Osika predicts AI-native challengers can move faster but says incumbents may retain trust and other advantages. The extent of disruption is uncertain and varies by industry.

On customer loyalty, Osika says some users favor brands while others switch based on cost and capability. His rough split is an impression rather than disclosed churn data. He argues established companies should study rapid organizational change as well as choose AI tools.

Stebbings advocates intense work and asks about balance. Osika favors sleep, exercise and some relaxation but believes a short period of very hard work can be appropriate for an ambitious goal. He says he focuses on impact rather than requiring identical hours from everyone; this is a leadership preference, not a guarantee of employee well-being.

He discusses the “keeper test”: whether a departure would make a leader try hard to retain someone. He says teams should consider both impact and cultural stability. The interview describes management’s intended method, not an independent review of employment practices.

Osika says he wants Lovable to improve quality and be more deliberate as it matures, rather than only pursue new ideas. He discusses building in Europe, finding talent and focusing on a core bottleneck. He says work on an adjacent open-source project diverted focus in hindsight, a company-specific lesson.

He identifies future bottlenecks in hiring technical product builders, improving the AI product and serving enterprise demand without losing focus on founders. He says Lovable intends to have an enterprise sales team, framed around understanding customers rather than a purely top-down sales model. The plan is not evidence of future execution.

Osika says hiring engineering leaders is difficult and admits he delegated too much in some hiring decisions. He wants to understand candidates’ motivation and expected outcomes more closely. These are reflections on his own process, not universal hiring rules.

Discussing co-founder Fabian, he says their contrasting styles can be productive: one favors simplicity and careful deliberation, while Osika is more inclined to try new ideas. He values candid discussion and humility in both business and marriage.

Asked what Lovable might look like at the end of 2026, he envisions a tool supporting the full product lifecycle, from an idea through customer communication and growth. That is a product aspiration rather than a description of completed functionality.

On benchmarks, Osika invokes Goodhart’s law: a metric can become less useful when teams optimize directly for it. He gives positive feedback clicks as an example of a metric that could be gamed. His critique does not show that all evaluation methods are unreliable.

In quick-fire questions, he gives an optimistic view of AI capability but also expects progress to plateau on some dimensions while advancing in others. He speculates about medical research applications; the interview does not establish clinical safety or evidence for new treatments.

Asked to choose among model companies at hypothetical valuations, Osika says he would buy xAI’s Grok and ultimately says he would short OpenAI, after initially mentioning Anthropic. He cites his impressions of team momentum and morale. These are speculative opinions, not verified workplace assessments or investment advice.

He thinks a leading future model could come from China and gives a rough 50-50 probability that Chinese models could become best. He says Lovable would need to assess data implications before using one. This is an uncertain prediction, not a measured probability or present purchasing decision.

He expects the best models to remain closed but says open models could be widely chosen for flexibility and ecosystem reasons. He also expresses interest in AI browsers and considers Figma a formidable potential competitor. These are forecasts and competitive views from a company executive.

Osika says he changed his mind about building an agent before models were ready: he now emphasizes getting a product into users’ hands and improving the whole experience from feedback. The described learning loop depends on customers actually finding lasting value.

Asked about job displacement, he worries rapid change and unemployment could create fear and social instability. He hopes societies can plan for transition and acknowledges potential shifts in knowledge work. Neither the number of jobs affected nor the success of any response is established in the interview.

He also worries that competition between powers could accelerate dangerous military uses of AI and calls for broader coordination. This is a risk scenario and policy view, not evidence that a particular outcome will occur.

Asked what success would look like, Osika imagines Lovable becoming a widely used interface between people and AI. He closes with that long-term company ambition; the market, adoption and revenue needed to reach it remain uncertain.