Aug 18, 2025, 9:54 AM ETFireside Chat, Interview

Lovable CEO, Anton Osika: The State of Foundation Models, Grok vs OpenAI, and Replit vs Bolt

Anton prefers investing in Grok over OpenAI or Anthropic, citing the "slope" and high morale of the Grok team, which is uniquely hiring "missionaries" for data curation (AI tutoring).
Anton suggests shorting OpenAI due to the recent internal "mess" and potential execution trade-offs in unifying five models into a single GPT-5 release.
There is a 50-50 probability that a leading model will originate from China, and Lovable plans to use Chinese models if they prove superior without compromising data privacy.
Anton believes Chinese model development is terrifying due to the speed of distillation and the frequency of high-quality releases (e.g., four new models weekly).
Anton predicts OpenAI may capture the consumer search market (next-gen Google) while Anthropic dominates the enterprise/developer space, though he acknowledges the outcome remains unknown.
Anton expects the best AI models to remain closed-source, though open ecosystems may win for flexibility.

Lovable's defensibility relies on becoming a "technical co-founder" that handles the entire product lifecycle (finance, admin, code, marketing), creating a high switching cost via accumulated user value.
Anton argues that "build mode" (shipping fast) is critical for survival, comparing AI startups to chickens shot from a cannon that must "flap fast" to gain traction.
The company aims to optimize for "tomorrow's model capabilities" rather than current ones, accepting higher compute costs now to iterate rapidly on future AI potential.
Lovable is moving beyond simple prototyping to integrate steps before code (validation) and after code (QA, marketing, growth) into a single unified workflow.
Anton views Figma as a potential competitor for the design-to-prototype phase but argues that pixel-perfect design slows down the AI-native product development cycle.
The platform is transitioning from an opinionated tool to an "agentic chain" that hyper-personalizes interactions based on deep context about specific users and projects.

Anton identifies "extreme trauma" or "extreme masochism" as desirable traits in hires, alongside a high "slope" indicating dynamic adaptability and excitement during interviews.
Lovable rejects the "founder mode" obsession for early-stage scaling, instead implementing a "protective layer" of organized managers to handle prioritization and order.
The company prioritizes hiring "generalists" with high impact potential over narrow specialists, as AI allows engineers to act as translation layers between product strategy and execution.
Anton advocates for an aggressive work culture ("996") for short-term bursts to achieve 10x impact, though he acknowledges the need for balance over 10-year horizons.
Lovable's hiring strategy includes a "Keeper System" to constantly evaluate if team members are optimally positioned to win, prioritizing impact over tenure.
The company plans to hire engineering leaders capable of innovating across multiple fronts, identifying this as the primary long-term bottleneck.

Lovable recently achieved $100M ARR in seven months, a milestone previously considered the gold standard for a two-year timeline.
The revenue breakdown is estimated at 80% complex application builders, 10% enterprise (product leaders prototyping), and 10% hobbyists.
Current unit economics are skewed by high compute costs; the goal is to shift to a subscription model where the majority of revenue is retained rather than passed to model providers.
Anton dismisses the idea that AI will commoditize model performance, predicting that high-value model providers will remain valuable and that Lovable's value accrual will increase.
The company plans to simplify token usage for users by abstracting the complex connections to model providers, potentially increasing margins through markup on token usage.

Anton regrets not focusing 100% on the core Lovable vision from day one, noting that the parallel open-source project (GPT Engineer) diluted maximal focus.
He believes university is generally a poor place to learn practical value creation, citing high opportunity costs and suggesting that "learning how to learn" is better achieved through work.
Anton advises against over-optimizing for metrics like "thumbs up" votes, as this incentivizes "hacking the metric" (e.g., telling fun jokes) rather than delivering genuine utility.
The company's long-term vision for 2026 is to become a "perfect co-founder" that manages the entire business stack from ideation to growth, including marketing and communication.
Anton remains unbothered by the potential for OpenAI or Anthropic to build competitors, betting on execution speed and superior user experience as the differentiators.

Anton expresses concern that hyper-competition between superpowers could lead to AI-enabled warfare or unintended catastrophic outcomes if human oversight is lacking.
He worries about societal panic regarding job displacement, arguing that humans often fail to define their goals in an era of rapid white-collar automation.
Lovable aims to build a "generational product" from Stockholm, challenging the notion that European startups lack the network, capital, and distribution of the US.
Anton suggests that large enterprises will face significant disruption as AI-native companies offer cheaper, faster alternatives for software development and data processing.
The company intends to launch an enterprise sales team to serve large organizations without adopting the traditional "wine and dine" sales culture.