Q&A with Palantir CTO Shyam Sankar on diffusing intelligence, safety, AI lessons, forward-deployed engineering, Anthropic vs. the US, AI religion, and more

Companies that build on top of AI models, rather than just those that create them, are more likely to succeed.

Palantir CTO Shyam Sankar discussed the AI industry's focus on "diffusion" versus "generation" in an interview. Sankar argued that while significant investment is going into creating AI models (the "power producers"), there is insufficient focus on how to integrate and utilize these models effectively within existing systems and workflows (the "machine tools"). He drew an analogy to the electrical revolution, where the impact came not just from generating electricity, but from building machines that could run on it. Sankar believes that companies focusing on the top-line of their operations will be more successful in adopting AI. He also expressed frustration with the AI labs' approach to safety, advocating for a systems engineering perspective rather than what he perceives as a refusal to address systemic risks. The interview covered Palantir's lessons learned from AI development, strategies for preserving "alpha" (unique advantage) over "beta" (market average), and the necessity of more AI models.

Sankar contends the AI industry is prioritizing the creation of intelligence over its practical application, akin to focusing solely on power generation without developing the machines to utilize that power. He suggests that success in AI adoption hinges on effectively integrating these capabilities into existing business processes, a concept he terms "diffusion," contrasting it with the current industry emphasis on merely developing more powerful models. This perspective implies a strategic shift for companies looking to leverage AI, moving beyond simply acquiring advanced models to concentrating on the engineering and operational frameworks needed for their widespread use.

The interview also touches on a perceived "AI religious war," where Sankar criticizes the AI labs' approach to safety, advocating for a systems engineering perspective that acknowledges and mitigates systemic risks. He differentiates Palantir's strategy by focusing on how to preserve a company's unique competitive edge ("alpha") rather than settling for average performance ("beta"), suggesting that true value from AI comes from novel applications and operational efficiencies that set a business apart, not just from adopting industry-standard tools.