North America consumption AI pattern — the deployment landscape in US/Canadian grocery, food delivery, QSR, and consumer nutrition (mid-2026)

NA-USA, NA-Canada

Content

This is the meta-unit for the North American consumer-facing AI landscape in grocery and food, mid-2026. It synthesises the deployment evidence across the related concrete units.

The pattern in seven observations

1. Conversational generative-AI is now mainstream at retailer-owned scale. Walmart Sparky (Jun 2025, deployed to all US app users) and Loblaw PC Express in ChatGPT (Feb 2026) anchor the retailer-owned end. Instacart Ask Instacart (May 2023, deployed to all US customers) anchors the third-party-platform end. Kroger × Google Gemini (Jan 2026) closes the Big Three US grocer × foundation-model matrix, with each major grocer tied to a different foundation-model vendor.

2. QSR voice-AI deployment has bifurcated into success and failure cases. Taco Bell × Omilia (890+ US drive-thrus, 38 states, Jul 2026) and Wendy’s FreshAI (~100 restaurants, 17 states, Jan 2025) are scaled production. McDonald’s × IBM AOT (100+ restaurants, ended Jul 2024) and DoorDash restaurant voice-ordering pilot (2-year run, ended April 2025) are the documented failures. Accent / dialect accuracy and B2B product-market fit are the failure modes.

3. Food-delivery platforms are agentic, not just conversational. Uber Eats AI Cart Assistant (2026) builds carts from text or recipe photos. DoorDash’s Ask DoorDash continues to evolve despite the voice-pilot shutdown. Instacart’s Smart Shop (Mar 2025) extends AI personalisation beyond search.

4. Consumer nutrition AI has three distinct sub-patterns:

   • Logging/computer-vision: MyFitnessPal × Cal AI (2026), with peer-reviewed 97% accuracy on food image recognition.
   • Biological-data personalisation: ZOE (microbiome + glucose + AI) and Levels Health (CGM-only).
   • Implicit-agentic grocery: Hungryroot (AI selects ~70% of customer baskets).
   • Noom, Samsung Food (the rebranded Whisk), and ChefGPT occupy adjacent space. None of these has surfaced as the dominant product.

5. Restaurant automation (robotics) is layered on top of consumer AI. Sweetgreen Infinite Kitchen (18+ new units in fiscal 2025; Spyce sold to Wonder Sept 2025 for $186.4M) is the largest in-store AI/robotics deployment in US fast-casual. Domino’s Dom (long-running voice) is the durability anchor.

6. Food-safety and processing AI is more deployed than the consumer narrative suggests. Tyson Foods computer vision spans bird counting, worker-safety PPE monitoring, and process automation across multiple US plants — invisible to the consumer.

7. The discontinuation rate is higher than vendor narratives suggest. McDonald’s × IBM, DoorDash voice, Whole Foods × Amazon Just Walk Out (Amazon divested 2024) all document that announced and even pilot-deployed AI surfaces in food frequently fail to scale.

Cross-cutting features of the NA pattern

• Foundation-model vendor diversity: Walmart↔OpenAI, Kroger↔Google, Loblaw↔OpenAI, Wendy’s↔Google, Sweetgreen↔in-house, Domino’s↔Nuance-era. No single dominant pairing.
• Standard-smartphone default: every consumer-facing surface requires a smartphone app; no deployment of voice-first consumer AI in agrifood at the consumer-side reaches the low-literacy or non-English-default population at scale. The only non-English-default capability in production is Wendy’s FreshAI Spanish (2024).
• Capital-intensive, vendor-mediated, retailer-led: parallels the Canadian retail AI pattern documented in canadian-retail-ai-pattern.md, but with US grocers reaching significantly larger customer footprints.
• Silent on consumer data sovereignty: every major deployment relies on vendor privacy policies, not public standards. US deployments additionally intersect with a patchwork state-level AI regulatory environment (California, Colorado, Illinois biometric laws) but agrifood-specific federal regulation is absent.
• Discontinuation rate: ~25–30% of high-profile announced AI deployments in this sector do not reach scaled production (estimated from the four documented discontinuations above: McDonald’s × IBM, DoorDash voice, Whole Foods × Just Walk Out, plus several smaller examples).

What this unit is doing in the taxonomy

This is the meta-unit for the US/Canadian consumer-facing AI landscape, parallel to canadian-retail-ai-pattern.md (which is Canada-only) and canadian-food-waste-ai-landscape.md (which is a vertical). It draws on the concrete deployment units to make a claim about the landscape.

Why it matters for talks

• Anchors any consumption-cell slide in the matrix with both breadth (retailer-owned, platform-owned, delivery, QSR, nutrition, processing) and depth (specific deployment numbers).
• Useful as a counter-narrative to vendor-fluent “AI everywhere in grocery” press: the deployment numbers and discontinuation cases together tell a more honest story than either alone.
• Documents the foundation-model-vendor diversity — a structural fact about the NA market that is not visible in any single concrete unit.

Critical context

• The deployment numbers cited are vendor-reported; independent audit data is sparse.
• The pattern is silent on what happens when AI gets things wrong: mis-pricing, mis-fulfilment, dietary errors, allergen mis-classification, voice ordering errors. No public incident-reporting mechanism is visible.
• The consumer-narrative bias in this reconnaissance is itself a taxonomy choice: back-of-house AI (Tyson) is documented, but the broader processing/supply-chain AI footprint (Blue Yonder, Loblaw forecasting, etc.) is in related units.
• US-specific regulatory vacuum: there is no US federal AI law specifically governing consumer-facing AI in food/grocery. State-level activity (California Bot Disclosure law) is partial.