The creation of effective evaluations ('evals') for complex, economically valuable work is the single largest barrier to automating most knowledge work, more so than improving the base reasoning of AI models.
AI-driven job displacement will be extremely fast and socially disruptive, creating a significant political problem in the near future.
Current base AI models possess sufficient reasoning for agentic work; the key challenge is teaching them tool use and company-specific knowledge through data-efficient fine-tuning.
AI is already superhuman at evaluating talent in text-based domains and is more effective than human hiring managers, making it irrational for companies not to use AI recommendations in hiring.
AI's potential as a manager may soon exceed its potential as an individual contributor, representing a paradigm shift in how we think about AI's role in the workplace.
▶The Primacy of 'Evals'Mar 2026
Foody's central thesis is that the largest remaining barrier to automating most knowledge work is not a lack of reasoning in base models, but the creation of effective evaluations. He argues that current benchmarks are too academic and that the focus must shift to evaluating performance on complex, economically valuable tasks, which he predicts will become a major job category itself.
This focus on 'evals' suggests a market opportunity in creating specialized, industry-specific AI assessment tools, potentially shifting value from foundational model creators to application-layer companies that can master evaluation and fine-tuning.
▶Imminent and Painful Labor Market DisruptionMar 2026
Foody holds a stark view on the speed and impact of AI-driven job displacement, describing it as happening 'very quickly' and creating a 'painful and a large political problem.' He sees this as an imminent reality, not a distant future, and predicts a hybrid marketplace where humans and AI agents compete directly for jobs.
Investors should consider the second-order effects of this predicted disruption, including political instability, the need for new social safety nets, and business models focused on managing the human-AI workforce transition.
▶AI as a Superior Talent Assessor and ManagerMar 2026
Foody claims AI is already superhuman at evaluating talent in text-based domains and that his company's models outperform human hiring managers. He extends this view to management, speculating that AI may soon be more effective as a manager than as an individual contributor, fundamentally changing organizational structures.
This theme challenges traditional HR and management paradigms, suggesting that companies that successfully integrate AI into hiring and management functions could gain a significant competitive advantage in talent acquisition and operational efficiency.
▶Pragmatic AI Application over Foundational ResearchMar 2026
Foody emphasizes practical application over chasing benchmark performance, arguing that base models are already capable enough for most agentic work. His focus is on the data-efficient customization of these models (via RFT) to learn company-specific knowledge and tool use, which he sees as the real bottleneck to value creation.
This perspective indicates that the next wave of AI innovation may be led by application-layer companies that can effectively integrate and customize existing models, rather than just the large labs building the models themselves.