Uber Eats Launches Cart Assistant: AI-Powered Agentic Shopping That Transforms Grocery Lists Into Carts
Key Takeaways
- ▸Uber Eats' Cart Assistant uses multi-prompt state graphs with LLMs and deterministic systems to convert natural language shopping requests into complete grocery carts automatically
- ▸The system shifts grocery shopping from a search-first model (intent → search → select items manually) to a cart-first model (intent → draft cart → review), reducing friction
- ▸Architecture separates concerns: LLMs handle intent interpretation and relevance judgment while deterministic APIs manage inventory retrieval, pricing, constraints, validation, and cart construction
Summary
Uber has launched Cart Assistant in beta on Uber Eats, a new feature that uses large language models and agentic workflows to automatically build grocery carts from natural language prompts or images. Rather than forcing shoppers to manually search for individual items, the system interprets user intent, retrieves relevant products from store inventory, judges relevance, applies pricing constraints, and generates a draft cart ready for review—shifting the paradigm from search-first to cart-first shopping.
The system is built on a multi-prompt state graph architecture where each step owns a focused piece of reasoning: intent planning, candidate retrieval and enrichment, semantic relevance judging, constraint enforcement, quantity conversion, validation, and cart aggregation. LLMs handle ambiguity and reasoning tasks while deterministic systems manage retrieval, pricing, eligibility checks, and schema validation. This hybrid approach allows shoppers to enter complex requests like "I want to cook pasta for two, plus vegan protein powder under $20 and paper towels" and receive a complete, contextually accurate cart.
Cart Assistant was featured at Uber's GO-GET product showcase, signaling the company's commitment to making grocery shopping more efficient through AI. The feature addresses a fundamental friction point in e-commerce: translating real-world intent—whether from a handwritten grocery list, recipe screenshot, or meal plan—into accurate, complete shopping carts without manual item-by-item search and selection.
- The feature handles complex, multi-constraint requests like recipes with ingredient lists, quantity adjustments for serving sizes, price limits, and product preferences
Editorial Opinion
Cart Assistant represents a practical maturation of agentic AI in commerce—moving beyond chatbots to systems that solve real coordination problems between natural language ambiguity and structured commerce constraints. By combining LLM reasoning with deterministic constraint systems, Uber demonstrates how AI agents work best when focused on narrow, high-value tasks rather than trying to be fully autonomous. This approach could become a template for other e-commerce platforms seeking to reduce shopping friction without sacrificing user control or system reliability.
