Marc Lore Uses AI and Blood Tests for Meals

According to Business Insider, Lore said on Semafor's podcast that for the past year, every meal he eats outside restaurants has been AI-directed: "When I'm not eating at a restaurant, every meal I eat is AI-directed." He said the system draws on his blood test results and biomarkers to plan his breakfasts, lunches, and dinners and has "kept me healthy." Lore said he hopes to roll out a public beta of the approach at Wonder food halls this fall. Business Insider reports Wonder has opened roughly 120 locations and has acquired Grubhub and Blue Apron; Wonder did not respond to Business Insider's request for comment. In earlier interviews, including a Fast Company Rapid Response conversation, Lore described feeding his blood work, Oura ring data, and glucose monitoring into an AI system that assigns him meals based on health goals and rated food preferences, and Nation's Restaurant News reported last year that Wonder was developing an app to visit customers' homes, run blood tests, and generate personalized weekly meal plans.

Systems that personalize diets from biomarkers typically combine lab or consumer blood-test data, feature engineering that maps biomarkers to dietary inputs, and a recommendation model that balances nutrition targets against a user's rated preferences, similar to how Lore describes the system learning from his meal ratings over time. Scaling this from one person to a food-hall product raises distinct challenges: integrating lab-standard biomarker data at scale, handling label scarcity for long-term health outcomes, and building recommendation models that stay both nutritionally sound and appealing across a large, diverse user base.

This is a useful case study in productizing a personalization model that currently exists as one founder's bespoke pipeline. Teams building similar systems should expect the hardest engineering work to be in menu-to-nutrition data mapping, real-time inventory alignment with meal plans, and quality-control layers that meet food-safety and regulatory expectations, not in the underlying recommendation algorithm itself.

Watch for whether Wonder's fall beta launches publicly and what cohort size or demographics it targets; whether the offering relies on clinical lab tests, at-home test kits, or continuous monitors, and what data-consent and governance practices are disclosed; and whether Wonder publishes any outcome metrics on biomarker changes or user adherence, which would be the clearest signal of whether the model generalizes beyond Lore's own routine.