“AI at the Checkout > AI is the Checkout” — Simon Taylor (Fintech Brainfood)
Why this is in the vault
The Walmart Sparky data is the cleanest publicly-disclosed proof of the merchant-owned-agent thesis to date — and it directly validates the Service-as-a-Software pattern from Rico’s piece earlier today (operator builds the agent → wins; aggregator tries to own the operator’s customer → loses). Worth filing as evidence-cluster doc on agentic commerce architecture. Bias note: forwarded by Tempo CEO; the article carves out a clean MPP/x402 lane that benefits Tempo’s positioning, which the founder explicitly flagged. Filing the bias separately so future readers see it.
⚠️ Bias disclosure
Forwarded route: Founder received this from the Tempo CEO. Tempo is the agentic-payments network the founder is building Phase 2 on (mainnet, $50 USDC test in flight per working-context).
The bias: Simon Taylor draws a clean distinction in the article: “MPP and x402 are designed for agents buying online resources (API access, etc.), ACP and UCP are about commerce, how do consumers find, select, and pay for goods and services online.” This carves out a separate, defensible addressable market for MPP/Tempo (agent-to-agent + API-economy payments) while pointing UCP at consumer commerce. Without this carve-out, Tempo would be competing directly with Stripe + UCP. With it, Tempo addresses a distinct layer.
Is the bias load-bearing? No — the technical distinction is real (agents buying API access ≠ consumers buying merch). But it’s a framing that benefits Tempo by preserving market definition. A more skeptical read would ask: as UCP matures, will it absorb agent-to-agent transaction patterns, leaving MPP as a vestigial protocol for niche use cases?
Note: Simon Taylor (@sytaylor) is the author of the Fintech Brainfood newsletter — independent fintech analyst with 68k followers, not a Tempo employee. He is widely-respected in the fintech-protocol-architecture space. The bias is in the framing, not the source.
The core argument
“The AI does not own the checkout. The merchant does.”
After two years of agent demos, chatbot checkouts, and “AI shopping” theatre, the market found its first real answer on April 24, 2026: Amazon, Meta, Microsoft, Salesforce, and Stripe joined the Universal Commerce Protocol (UCP) Tech Council — sitting alongside Google, Shopify, Etsy, Target, and Wayfair. Every hyperscaler. Every major commerce platform. Every payments network. Every relevant marketplace. First time the agentic commerce category has agreed on anything.
The metaphor: UCP is the shipping container for agentic commerce. Before containers, global trade existed but every handoff was bespoke. The container standardized the interface — it didn’t own the goods, didn’t replace the merchant. It made the system interoperable. UCP does the same for agent-to-merchant interactions.
The four UCP layers
- Discovery: Agents query merchant
/.well-known/ucpmanifest to discover capabilities (loyalty programs, pre-orders, catalogs). - Negotiation: AI and merchant backend negotiate real-time pricing, tax, complex discount stacking via standardized APIs.
- Transaction: Secure API-driven checkout sessions, often using AP2 (Agent Payments Protocol) for verified one-tap payments (Google Pay, Shop Pay).
- Fulfillment & Settlement: UCP orchestrates logistics + provides post-purchase status updates back to the agent.
UCP is not a payment protocol. Payment still goes through traditional rails (cards, ACH, stablecoins). UCP coordinates everything around the payment.
The agentic commerce stack (per Simon’s earlier “agentic payments map”)
| Layer | Question answered | Protocols |
|---|---|---|
| Agent-to-agent communication | How do agents talk to each other? | A2A |
| Mandate | Does the agent have authority to pay? | AP2 |
| Transaction (commerce) | What’s being bought, how does checkout work? | UCP, ACP |
| Transaction (API economy) | Agents buying online resources / API access | MPP, x402 |
| Payment rails | Money movement | Cards, ACH, stablecoins |
This is where the Tempo bias sits. The MPP / x402 placement as a distinct “transaction layer for API economy” is what carves out Tempo’s defensible market. Otherwise MPP would compete with UCP for transaction coordination.
UCP vs ACP
- UCP (Google + Shopify): Started with full commerce flow, worked inward. Merchant owns the cart, owns the checkout, remains merchant of record. Agent is the discovery surface, not the storefront.
- ACP (OpenAI): Started with checkout (ChatGPT Instant Checkout) and worked outward. Agent owned the relationship.
The two specs are converging. ACP moved toward UCP’s merchant-owned model. UCP’s architectural advantage: deliberately layered like TCP/IP. Core checkout primitives at the bottom, capabilities above (Catalog, Orders, Checkout), extensions on top (loyalty, fulfillment, subscriptions). PSPs, commerce platforms, hyperscalers can extend without permission. That layered architecture is why coalitions trusted it early enough to converge.
The Walmart receipt (load-bearing evidence)
Walmart ran the experiment both ways:
Path A: AI owns the checkout (ChatGPT Instant Checkout, ACP-style)
- Walmart tested 200,000 SKUs through ChatGPT Instant Checkout
- Conversion was 1/3 of click-out rates (Daniel Danker, EVP AI Acceleration, told WIRED)
- Called the experience “unsatisfying”
- Two weeks later, OpenAI announced the closure of Instant Checkout
Path B: Merchant owns the checkout, AI is the discovery + assistance surface (UCP-style)
- Walmart embedded Sparky (their assistant) in their app + later as a ChatGPT app
- Cart, login, checkout stayed on Walmart.com
- Half of Walmart’s app users have engaged with Sparky
- Sparky users have AOV +35% over non-Sparky users
- Sparky-inside-ChatGPT converts at 70% of Walmart.com direct rates — more than 2x ChatGPT Instant Checkout’s run rate
Other merchant-owned-AI receipts:
- Tatcha (Unilever luxury skincare): on-site AI assistant attributes 11.4% of total site revenue, conversion 3x site average, AOV +38%
- Microsoft Copilot Checkout: 53% more purchases within 30 minutes when shopping intent is present
The pattern is consistent across every dataset where the merchant owns the AI: AI by the merchant works. AI by the aggregator doesn’t yet.
The mainstream-isn’t-here-yet counter-evidence
October 2025 peer-reviewed study (Maximilian Kaiser, University of Hamburg + Christian Schulze, Frankfurt School of Finance) — first peer-reviewed study of LLM e-commerce traffic. 12 months of first-party data from 973 e-commerce sites, 50,000 ChatGPT-referred transactions vs 164M traditional transactions:
- ChatGPT referrals underperformed every traditional channel except paid social
- Affiliate links converted 86% better than ChatGPT
- Organic search converted ~13% better than ChatGPT
- ChatGPT accounted for 0.2% of total traffic across the dataset
- Authors concluded parity with organic search inside a year is “unlikely”
But the early-adopter signals are growing fast:
- Adobe: AI-driven retail traffic up 693% YoY for 2025 holiday season; AI referrals converting 31% better than other channels (vs. 9% worse just three months earlier)
- Shopify Q4 2026 earnings call: AI-attributed orders grew 15x since start of 2025
- Panxo: ChatGPT converts at 11.4% — beats direct, paid search, organic, email, display, social
The synthesis: mainstream not here yet. Off a tiny base, AI commerce is growing fast. Merchant-owned implementations (Walmart Sparky, Tatcha) materially outperform aggregator-owned ones (ChatGPT Instant Checkout, RIP).
The strategic prescription
“The merchants who get this will run twenty UCP experiments in 2026. The merchants who don’t will run one in 2027, when the data is already in, and the option is already expensive.”
Simon’s actionable list:
- Be discoverable to agents (AEO — Agent Experience Optimization, basically SEO for agents)
- Build agent-friendly checkouts (UCP-ready merchant backend)
- Be ready to communicate (merchant-side AI, like Sparky)
- Optimize that experience over time
The merchants who win run AI on both sides: consumer agent for discovery + merchant agent for conversion. UCP is what lets them talk.
Mapping against Ray Data Co
Three load-bearing connections to active RDCO positioning:
1. Validates the Service-as-a-Software thesis 2026-05-03-heyrico-service-as-a-software-shift in a different vertical
Same pattern, different industry. Operator builds the agent → wins. Aggregator (OpenAI) tries to own the operator’s customer relationship → loses. Walmart’s Sparky data is the consumer-commerce proof of what Rico argued in services. RDCO’s MAC + Client Reporting bets follow the same logic for analytics-engineering: build agent-tooling FOR shops they OWN, don’t try to be the aggregator routing customers through.
2. MPP/Tempo Phase 2 positioning gets cleaner
If RDCO ever monetizes Ray’s outputs (other 1-person cos paying per-decision-surface, or a client-reporting agent charging per-report), MPP rails fit better than UCP rails because that’s agent-to-API-economy not consumer-commerce. Doesn’t change Phase 2 plan but confirms the addressable-lane positioning. Watch for whether UCP eventually absorbs agent-to-agent patterns.
3. Squarely partial impact via AEO
App Store algorithm IS a discovery surface for consumer agents. If ChatGPT/Gemini/Claude start fielding “find me a daily logic puzzle game” queries, Squarely needs to be discoverable. Adjacent to Simon’s AEO point. Worth a single line in Squarely STRATEGY.md but not a re-prioritization.
4. Sanity Check material (Walmart Sparky as macro-bookend)
If founder eventually writes the Service-as-a-Software piece grounded in his own client-reporting receipts (per the conversation earlier today on operator-log positioning), Walmart Sparky is the macro-bookend that connects his lived experience to the broader thesis. Founder’s operator data is primary source; Walmart receipts + Rico framing become the bridge.
Open questions for founder
- Should the agentic-commerce architecture map (UCP/ACP/MPP/x402/AP2/A2A layers) get its own concept doc? Reference doc that future bets can map against without re-deriving the stack each time. ~30 min to draft, useful when next agentic-commerce conversation surfaces.
- AEO for Squarely — worth a single-paragraph addition to Squarely STRATEGY.md? “Discoverable in consumer agent surfaces” is adjacent to App Store ASO but not identical. Low cost to capture as a strategic dimension.
Numbers worth remembering
- April 24, 2026: UCP Tech Council expansion adds Amazon, Meta, Microsoft, Salesforce, Stripe (alongside Google, Shopify, Etsy, Target, Wayfair)
- McKinsey: $3-5T agentic commerce category by 2030
- Morgan Stanley: $190-385B US e-commerce by 2030 (10-20% of online retail)
- Hamburg/Frankfurt peer-reviewed study (Oct 2025): ChatGPT 0.2% of total traffic, parity with organic search “unlikely” within a year
- Adobe 2025 holiday: AI-driven retail traffic up 693% YoY; AI referrals converting 31% better than other channels
- Walmart Sparky: 50% of app users engaged; AOV +35%; ChatGPT-Sparky converts at 70% of Walmart.com direct (2x ChatGPT Instant Checkout)
- Tatcha: AI assistant = 11.4% of site revenue, conversion 3x site avg, AOV +38%
- Microsoft Copilot Checkout: +53% purchases within 30 minutes when shopping intent present
- Apple Pay benchmark: ~14% of US online payments today (started 2014); table stakes on mobile
Related
- 2026-05-03-heyrico-service-as-a-software-shift — Rico’s Service-as-a-Software thesis; Walmart Sparky data is the cleanest consumer-commerce proof
- 2026-05-03-yc-build-company-with-ai-from-ground-up — YC’s four-frame articulation; “AI as OS” + “queryable company” tie to UCP’s manifest-discovery pattern
- 2026-05-02-moonshots-ep252-google-anthropic-gpt55-cloud — Wissner-Gross’s per-token economics is the underlying gradient; UCP is the merchant-side response to per-token agent dominance
- 2026-04-30-trevin-chow-orchestration-thesis — orchestration layer thinking, adjacent to “agent IS the operator” frame UCP enables
~/.claude/state/working-context.md— MPP/Tempo Phase 1 testnet wallet + Phase 2 mainnet plan; UCP/MPP carve-out is the strategic context- ../01-projects/squarely-puzzles/STRATEGY — partial relevance via AEO/agent-discoverability dimension
- ../01-projects/positioning/STRATEGY — agent-deployer positioning that Walmart Sparky pattern validates
Source-fidelity notes
Article fetched via xmcp.getPostsById with article field expanded. Author @sytaylor (Simon Taylor, Fintech Brainfood, 68.7k followers, independent fintech analyst). Engagement: 11,385 impressions, 35 likes, 44 bookmarks, 3 retweets, 2 replies, 2 quotes. Bookmark/like ratio of 1.26x — moderate save-for-later. Article cites: Walmart Q4 FY26 earnings, Wired Walmart-OpenAI piece, Adobe AI-driven traffic blog, McKinsey retail insights, Morgan Stanley reports, Panxo data, Capital One Apple Pay stats, Hamburg/Frankfurt SSRN study, Walmart’s Sparky launch coverage, UCP newsfilecorp release, Target+Gemini Business Insider piece, Shopify engineering UCP post, Fintech Brainfood “Ramp cracked AI” + “agentic payments map” + “checkout is dead” prior pieces.
Direct quotes ≤15 words throughout, in quotation marks. Source URL + article URL preserved.