Nitski argues that open-source models won't cannibalize data providers because data is most valuable at the frontier of model performance. He believes there is no enterprise AI ROI problem, only a period of exploration, and that specialized models tailored to each company's priorities will emerge. For founders, balancing AI performance against token costs depends on the use case, with token spend seen as a valuable COGS-like investment. Product teams must simplify surface area and focus on judgment over tools, as AI reduces skill barriers. The boom in AI services and forward-deployed engineers is here to stay due to concentrated expertise in San Francisco. Mercor aims to diversify beyond frontier labs by building self-serve products and expanding into enterprise agent arms and robotics data. Robotics progress will be gradual, like Waymo's scaling, rather than a sudden 'ChatGPT moment.'