Visa put 22 AI models from 12 labs through 15 shopping challenges and they failed 42.8% of test runs, Visa said on 5 November; in 54% of the failed runs the AI still made a purchase (our piece).
In the same week, Anthropic gave Claude read-only access to users’ bank and card accounts through Plaid (our piece). Amazon took its own shopping agent into text messages after blocking Meta’s Muse from querying its site in September, as TechCrunch reported (our piece). By our reading, each of the three checkpoints an AI shopping agent meets this week was set by an established firm.
Business Pill · Agentic Commerce
An evergreen Business Engineer explainer on agentic commerce. It teaches the general idea only and does not discuss this story.
The key insight: This week Visa said 22 AI models failed 42.8% of its shopping tests, Claude got read-only bank access through Plaid, and Amazon put its own shopping agent in text messages after blocking Meta’s Muse in September; by our count, all three checkpoints were set by established firms.
The card network wrote the test. Visa’s benchmark requires an AI to discover every customer requirement and then make a correct buy or no-buy decision for a run to pass. Its release says: “If nothing fits the customer’s requirements, the AI should opt not to buy.” By our arithmetic, about 23% of all runs, roughly 229 of 990, ended in a purchase after a failed test (our piece).
The bank link is read-only. Anthropic’s Personal Finance, a US beta on the Pro, Max and Ultra plans, connects accounts through Plaid, and Claude’s launch post says “Claude can’t move, manage, or invest your money.” Users sign in to their bank through Plaid, “so Claude doesn’t ask for or see your bank password” (our piece).
The store runs its own agent. Amazon’s third-quarter release says Alexa for Shopping “was used by 400 million+ customers in the last 12 months”. On the 29 October call, Andy Jassy said: “though we anticipate working with horizontal agents we believe Alexa for shopping is going to be the primary shopping agent customers use to shop Amazon”, per a machine transcript (our piece).
Walmart’s agentic access terms require a human to approve every checkout by an outside AI agent unless it shops through Walmart’s own agent, Sparky (our piece).
The three checkpoints an AI shopping agent met this week: card network, bank-data link, store. Source: Visa, Anthropic, Amazon and TechCrunch via our coverage, 5-7 November 2026. Count and classification: ours.
Our Number
By our count, 3 of the 3 checkpoints in this week’s agent-commerce news, the card network, the bank-data link and the store, were set by established firms rather than by the AI labs whose agents shop. The classification is ours. Visa’s 42.8% failure rate, Anthropic’s read-only access and Amazon’s 400 million+ figure are the companies’ own.
The Structural Read
As we read these pieces together, AI agents this week could search, compare and advise, but the steps that move money or complete a sale ran through firms that already controlled payment, bank data or the storefront. This is a description of where the checkpoints sit, not a finding that any firm is blocking competition.
The Structural Read
As we read our own coverage, the AI agent is the newcomer at each step this week: the card network wrote the test it is scored on, the bank-data link lets it read but not move money, and the store runs its own agent.
As we read it, the established firms are also building their own agent products: Amazon’s Alexa for Shopping, which its release says 400 million+ customers used, and Walmart’s Sparky, per its access terms.
As we read it, this is a map of where the checkpoints sit this week, not a finding that any firm is blocking competition.
Claude, launch post on X
“Claude can’t move, manage, or invest your money.”
Three Implications
FOR AI LABS BUILDING SHOPPING AGENTS Visa’s first round failed 42.8% of runs, and 54% of those failures still bought, per Visa; the benchmark is the card network’s own.
FOR BANKS AND DATA AGGREGATORS Claude’s Personal Finance reads accounts through Plaid and cannot move money, per Anthropic.
FOR RETAILERSAmazon runs its own agent and Walmart’s terms require a human to approve outside agents’ checkouts unless they use Sparky, per our earlier pieces.
The analysis says: “The surface, not the model, governs autonomy.”
As we read it, this week’s checkpoints for shopping agents, the test, the bank link and the storefront, are all surfaces around the models; the analysis does not discuss these stories.
What Is Not Established
Our pieces do not establish Amazon’s plans for outside agents on its site, Anthropic’s plans for payment actions in Personal Finance, or how agents score on Visa’s benchmark today. Visa’s benchmark measures its own 15 challenges, and Amazon’s 400 million+ and Muse’s activation figures measure different things.
This Week In The AI Supercycle
The Great Convergence of Capital, Power, and Geopolitics
This week’s issue: for two years the AI build-out was priced as a supply story, and last week as a credit story. This week it was priced as a permission story, and the answer arrived as a cost, not a ban.
AI shopping agents failed 42.8% of Visa’s tests, Claude’s bank access arrived read-only, and Amazon put its own agent in text messages after blocking Muse in September, so by our count all three checkpoints this week sat with established firms.
A note on sourcing. This analysis draws only on our own earlier coverage, linked in the text; the count and classification of checkpoints are ours, and the 23% share is our arithmetic from Visa’s figures.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
