Consumer-AI and hospitality AI as labour-displacement negative-result — Sweetgreen + McDonald's + Kroger + DoorDash discontinued cluster

NA-US (multi-state + multi-deployment)

Content

The consumer-AI and hospitality AI cell is the corpus’s most-evidenced labour-displacement negative-result layer. Six deployments anchor the cell: Sweetgreen Infinite Kitchen (in-store robotic salad makeline, FY2025 deployment + September 2025 Spyce divestment to Wonder for $186.4M), Wendy’s FreshAI × Google Cloud (~100 restaurants / 17 states / January 2025 + 86% completion-without-human-intervention vendor-reported metric), Taco Bell × Omilia (890+ drive-thru voice AI units), McDonald’s × IBM AOT (100+ restaurants in NA, discontinued June 2024), Kroger × Ocado CFC closures (Frederick MD + Goodyear AZ + third unreported, discontinued November 2025 with the “bet on robotics went too far” corporate post-mortem), DoorDash voice AI discontinued. Four high-profile US consumer-AI / hospitality robotics deployments walked back in an 18-month window. This is structurally novel as an industry pattern.

This is a consolidated cell anchor unit — distinct from single-vendor units. The cell-level finding is the substantive contribution: consumer-AI and hospitality AI has the most-evidenced labour-displacement-doesn’t-happen-yet layer in the corpus, and that pattern is itself the substantive analytical move.

Sweetgreen Infinite Kitchen

Per units/sweetgreen-infinite-kitchen.md:

• What it is: Sweetgreen’s automated bowl/salad makeline; uses robotics + AI/ML to assemble customer orders. System originated from Sweetgreen’s 2021 acquisition of Spyce (MIT-spinout automated kitchen company).
• Deployment scale: 18 of 37 net new restaurant openings fiscal 2025 + 7-9 new units + 2-4 retrofits in earlier 2024 plans. At least 25 new Infinite Kitchens planned for 2025 per Q3 2025 earnings release.
• September 2025 divestment: Sweetgreen sold the Spyce business to Wonder for $186.4M, while retaining the right to deploy the technology in its own restaurants “with even greater [speed]”.
• Critical context named in unit: “Restaurant automation has a labour displacement dimension — Sweetgreen has not publicly committed to a no-layoffs policy tied to Infinite Kitchen deployment. The September 2025 sale to Wonder is itself a strategic signal: Sweetgreen is exiting the technology business to focus on the restaurant business, while licensing back the technology. This is a common pattern but worth surfacing.”
• Labour-displacement framing: Sweetgreen has not publicly committed to a no-layoffs policy tied to Infinite Kitchen deployment. The makeline is the largest deployed AI/robotics system in US fast-casual restaurant consumption as of mid-2026.

Wendy’s FreshAI × Google Cloud

Per units/wendys-freshai-google.md:

• Deployment scale: ~100 restaurants across 17 states running FreshAI in production (January 2025).
• Vendor metrics: 86% average order-completion rate without human intervention (vendor number); ~99% accuracy (Wendy’s broader definition: any AI-started order submitted to POS); one test site showed service times 22 seconds faster than Columbus market average.
• Spanish language capability: Added in 2024. The only public North American QSR voice-AI deployment that explicitly handles a non-English language in production as of mid-2026.
• Wendy’s operates over 7,000 restaurants globally, so 100 deployed = ~1.4% of footprint. The production reality gap: 100 of 7,000+ restaurants is small. Public framing tends to over-state scaling.
• Voice data governance: Wendy’s CIO frames FreshAI as a platform intended to scale to mobile apps, kiosks, and smart devices, beyond drive-thru. Voice data flows to Google Cloud.

Taco Bell × Omilia

Per units/taco-bell-omilia-drive-thru-voice-ai.md:

• Deployment scale: 890+ drive-thru voice AI units — the largest voice AI drive-thru deployment by unit count in the corpus.
• Smaller per-unit deployment than Wendy’s FreshAI but with stronger unit-count scale; public metrics are limited.

McDonald’s × IBM AOT (DISCONTINUED June 2024)

Per units/mcdonalds-ibm-ai-drive-thru-discontinued.md:

• Deployment scale: 100+ McDonald’s US drive-thru locations starting 2021.
• Timeline: 2021 McDonald’s acquires Apprente (voice-AI startup) → renames McD Tech Labs → December 2021 IBM acquires McD Tech Labs → 2021-2024 AI drive-thru pilot in 100+ restaurants → June 2024 McDonald’s announces end of partnership with IBM; voice ordering AI shut off at all pilot locations “no later than July 26, 2024.”
• Reasons cited in coverage: Order accuracy at noisy drive-thru (SOTA voice models at 2021-2024 did not yet match noise tolerance); edge cases (special instructions, substitutions) slowed service; customer experience degradation when AI mis-ordered; workers’ and franchisees’ reports of cost-to-serve issues when AI failed.
• Workforce framing named in unit: “QSR AI drive-thru has been framed as a labour story by some media (displacement concerns) and as an efficiency story by others (faster service, fewer mistakes). The pilot ending complicates the labour-displacement concern by demonstrating that the technology itself wasn’t ready — though it does not resolve the policy question of what happens if/when it works. Worker / labour union concerns were raised during the pilot period; the discontinuation may have been partly tactical (cost / accuracy) rather than a labour settlement.”
• Strategic refocusing: McDonald’s did not declare AI-driven ordering a dead-end — they ended the IBM partnership specifically, with reports of subsequent partnerships with Google for AI drive-thru replacing the IBM role. The discontinuation was tactical, not strategic.

Kroger × Ocado CFC closures (DISCONTINUED November 2025)

Per units/kroger-ocado-cfc-closures-discontinued.md:

• Deployment scale: Multiple CFC sites — Monroe OH, Groveland FL, Dallas TX, Frederick MD, Forest Park GA, Goodyear AZ.
• Closure announcement (November 2025): Kroger announced closure of three CFC sites including Frederick MD and Goodyear AZ; Kroger publicly stated the “bet on robotics went too far” (per Grocery Dive November 19, 2025). One of the highest-profile US grocer robotics-deployment walk-backs in the agrifood sector.
• Why this matters: CapEx-heavy physical infrastructure investment ($hundreds of millions); Kroger’s post-mortem acknowledgement is unusually candid for a grocer. “The Kroger × Ocado closure is structurally similar to Sweetgreen’s 2024 Spyce sell-off to Wonder (Sweetgreen divested its robotic-kitchen subsidiary). Two distinct US agrifood companies walking back robotics investments in the same 18-24 months.”
• Workforce framing: “CFCs rely on automation rather than picker labour at fulfilment centres; the closure narrative reverses this — automation was too capital-intensive, not just too labour-displacing.”

DoorDash voice AI discontinued

Per units/doordash-ai-voice-ordering-discontinued.md:

• Deployment scale + discontinuation: DoorDash voice AI discontinued. Specific deployment scale + discontinuation date + reasons cited in coverage not surfaced in the existing unit (substantive gap to be closed in deep phase).

The consolidated cell finding — labour-displacement-doesn’t-happen-yet

Four high-profile US consumer-AI / hospitality robotics deployments walked back in an 18-month window. McDonald’s × IBM ended June 2024. DoorDash voice AI discontinued. Kroger × Ocado closures November 2025. Sweetgreen Spyce sold September 2025 (with deployment rights retained). The pattern is structurally novel:

The labour-displacement dimension is constrained by technology readiness + capital economics. Where the technology wasn’t ready, labour displacement didn’t happen — even where the intent (cost reduction, automation, scaling) was explicit.

The empirical record is more nuanced than the dominant vendor / investor pitch. The labour-displacement question is answered differently depending on whether the deployment walked back or stayed.

The labour-displacement framing in working deployments

In working consumer-AI / hospitality deployments, the labour-displacement framing is:

• Sweetgreen Infinite Kitchen — labour-displacement dimension named; Sweetgreen has not publicly committed to a no-layoffs policy. The makeline replaces the salad-assembly line; the labour-reduction dimension is direct.
• Wendy’s FreshAI 86%-completion-without-human-intervention — the most-specific public labour-automation-rate metric on a voice-AI deployment. The metric is vendor-reported (Wendy’s own methodology) but methodologically specific. The labour-automation-rate is the headline number for any labour-spine talk about voice AI drive-thru.
• Taco Bell × Omilia 890+ units — public metrics are limited; the deployment scale is the substantive finding.

In discontinued consumer-AI / hospitality deployments, the labour-displacement framing is:

• Constrained by technology readiness (McDonald’s × IBM accuracy problems; Kroger × Ocado capital economics).
• The deployment walked back rather than sustained — labour displacement didn’t happen because the deployment itself didn’t sustain.
• Strategic refocusing rather than labour settlement — McDonald’s moved to Google; Sweetgreen divested to Wonder; the technology was deferred not rejected.

Three substantive findings from this consolidated cell anchor

Finding 1: The discontinued cluster is the corpus’s most-evidenced labour-displacement-doesn’t-happen-yet layer. Four high-profile US consumer-AI / hospitality robotics deployments walked back in an 18-month window. The pattern is structurally novel.

Finding 2: The labour-automation-rate metric (Wendy’s 86%-completion-without-human-intervention) is the most-specific public labour-side metric on a voice-AI deployment. The metric is vendor-reported but methodologically specific. The labour-spine talk about voice AI drive-thru should foreground the metric.

Finding 3: Working vs discontinued cluster have structurally different labour-displacement answers. In working deployments, labour-displacement is real and bounded (Sweetgreen has not committed to no-layoffs; Wendy’s 86% completion-without-human-intervention is the headline number). In discontinued deployments, labour-displacement was constrained by technology readiness + capital economics.

What this unit is doing in the taxonomy

Anchors the consumer-AI / hospitality AI × labour-displacement negative-result cell — a consolidated cell anchor unit. Distinct from:

• Single-vendor units (sweetgreen-infinite-kitchen.md, wendys-freshai-google.md, taco-bell-omilia-drive-thru-voice-ai.md, mcdonalds-ibm-ai-drive-thru-discontinued.md, kroger-ocado-cfc-closures-discontinued.md, doordash-ai-voice-ordering-discontinued.md) — this unit is the labour-side consolidation of all six.
• Other consumer-side units (albertsons-afresh-fresh-suite.md, loblaw-blue-yonder-forecasting.md, walmart-sparky-ai-shopping-assistant.md, instacart-cart-assistant.md) — distinct cells.
• UFCW / Restaurant Workers Union / SEIU / UNITE HERE formal position on consumer-AI labour — not surfaced as primary source; G-3XX (new).

Why it matters for talks

• The discontinued cluster is the substantive analytical move. Per the labour-spine scan: “Where the technology wasn’t ready, labour displacement didn’t happen — even where the intent was explicit. The empirical record is more nuanced than the dominant vendor / investor pitch.”
• Wendy’s 86%-completion-without-human-intervention is the most-specific public labour-automation-rate metric on a voice-AI deployment; the labour-spine talk should foreground it.
• Strategic refocusing pattern (Sweetgreen Spyce → Wonder; McDonald’s IBM → Google) is a teachable pattern on AI deployment in restaurants; the technology is deferred not rejected.
• The labour-spine talk should foreground the working vs discontinued cluster distinction — labour-displacement is real in working deployments and constrained in discontinued deployments; both framings should be held without collapsing either.

Critical context

• All deployment evidence is from vendor sources. Wendy’s 86%-completion-without-human-intervention is Wendy’s own metric; Taco Bell public metrics are limited; Sweetgreen deployment numbers are vendor-reported; Kroger post-mortem is unusually candid corporate acknowledgment.
• DoorDash voice AI discontinued specific deployment scale + discontinuation date + reasons cited in coverage not surfaced in the existing unit; substantive gap to be closed in deep phase.
• Front-of-house worker voice on AI deployment is not surfaced as primary source; G-3XX (new).
• Restaurant Workers Union / SEIU / UNITE HERE formal position on consumer-AI labour is not surfaced as primary source; G-3XX (new).
• Back-of-house worker displacement quantification in working consumer-AI deployments is not surfaced; G-3XX (new).
• Voice AI 2026-vintage viability — Whisper-class accuracy improvements since 2024 discontinued cluster; whether voice AI for noisy drive-thru is now viable at commercial scale; G-3XX (new).