20VC: Mercor CPO on Revenue Concentration from Frontier Labs | Why Large Enterprise is Scared to Partner with Frontier Labs | Why Small Specialised Models is the Future with Osvald Nitski

This section explores how traditional closed frontier models are being challenged by open-source models. Osvald Nitski explains that open models are not yet cannibalizing the core business of companies like Mercor because demand is still focused on the unique capabilities provided by proprietary, closed models. The growth in demand for specialized, enterprise-specific models is driven by latent needs for long-horizon tasks and automation workflows that open models haven't yet addressed. The discussion emphasizes that the market for data supporting these advanced workflows is expanding, and companies continue to innovate in both open and closed model domains.

The conversation delves into why large enterprises are cautious about sharing data with frontier model providers. Sectors with core, sensitive operations like legal or medical services exhibit higher hesitation, particularly with Chinese open-source models. Osvald underscores that open weights models can run inference in multiple locations, offering more control, but enterprises remain wary due to data privacy and security concerns. The challenge remains in balancing proprietary data use with the need for open models, and educating enterprises about safe deployment options.

This section discusses the current landscape of AI investment and spending. Osvald reassures that there's no major ROI problem currently because the market is in an exploratory phase with high patience for experimentation. He highlights different approaches for managing AI budgets, including aggressive investment by hyper-growth companies and more measured spend by large corporations. The conversation emphasizes that as the field matures, better tools and understanding will enable more precise ROI calculations. Companies like Salesforce spending hundreds of millions annually on AI illustrates ongoing large-scale investments, with expectations that these percentages will increase over time.

The discussion addresses how Mercor manages its rapid growth, building more products quickly while trying to keep operations streamlined. Osvald talks about the difficulty in balancing product surface area—expanding features versus simplifying—and the importance of reducing complexity to ensure scalability. He highlights that product teams are evolving to include higher ratios of product managers to engineers, focusing on understanding user needs and business impact, especially in a high-demand, fast-paced environment.

The episode considers what competency areas are crucial for product managers in an AI context. Osvald notes a shift towards higher-level strategic skills, judgment, and business impact awareness, with less emphasis on tool mastery. He also discusses talent acquisition, favoring more senior, business-savvy candidates, and the high agency and ownership levels expected in top talent. Furthermore, the conversation emphasizes the significant challenges of hiring, retaining high-quality talent, and the importance of company culture. Notably, Aftersales and high agency individuals are viewed as key to long-term success.

Osvald expresses excitement about the potential for robotics to reach a ChatGPT-like 'moment', where robots become more integrated and useful in practical tasks, especially in autonomous driving and home automation. He compares the current state to the evolution of self-driving cars, predicting a similar breakthrough will occur, but likely in a gradual, 'Waymo-robotaxi' style adoption, emphasizing that these technologies are progressing faster in certain markets like San Francisco.

The segment highlights the rising importance of AI in cybersecurity, where adversarial, evolving data types drive the demand for sophisticated defensive and offensive AI models. Osvald notes that cybersecurity presents continuous challenges due to its adversarial nature, requiring dynamic data types, and it's an area poised for significant growth. He emphasizes the ongoing arms race and the need for real-time, agile AI solutions to improve defensive capabilities.

The conversation touches on the difficulty of recruiting and retaining top AI talent in Europe, especially in Estonia, due to talent scarcity and high costs. Osvald points out that talent war is fierce globally, but being part of a fast-growing company can ease hiring challenges. He advocates for focusing on hiring individuals with high agency and ownership, who can learn quickly and deliver impact, rather than solely relying on experience from big tech.

Finally, Osvald emphasizes the importance of maintaining a startup mentality—paranoia, agility, and high-performance culture—as Mercor expands. He shares insights on team building, including outdoor events and fostering high agency. The discussion underlines the value of culture, rapid iteration, and thoughtful hiring in sustaining hyper-growth and innovation.