Visa tested 22 AI models as shoppers, and they failed 42.8% of the time
THE SO WHAT
Visa’s benchmark showing 22 AI models failing 42.8% of shopping tasks is a hard constraint on near-term agentic commerce. If you’re planning checkout or purchasing agents, scope them as copilots with human confirmation, not fully autonomous buyers, and instrument failure modes aggressively.
READ THE SOURCE
MORE FROM THE WIRE
Applied AIPartnering with Edelbyte: Data Infrastructure That Keeps Up With AI's Curiosity
If agents are going to hammer your internal systems with orders of magnitude more queries than humans ever did, the real constraint becomes data infra cost and latency, not model quality. Teams building agents should be budgeting for a new class of OLAP-style infra like Edelbyte’s, not assuming the existing warehouse can take the load.
Applied AISoundCloud lets ElevenMusic users publish AI-made songs directly
Direct ElevenMusic → SoundCloud publishing turns AI tracks into first-class inventory, not side-loaded experiments. Labels and rights holders now have to treat AI catalogs as distribution-native, not fringe uploads.
Applied AIAI Selloff Has Room to Go on Mechanical Selling, Says Academy’s Tchir
If the AI trade unwinds further on mechanical selling while compute capex slows, late-stage infra and tooling startups will feel it first in fundraising and logo velocity. Operators should assume a bumpier demand curve and prioritize cash runway and real unit economics over “AI multiple” stories.
Applied AIAnthropic launches Personal Finance in Claude, letting users connect bank and credit card accounts via Plaid with read-only access to analyze spending trends
Claude moving into Plaid-connected personal finance makes LLMs a front-end for transaction data, not just chat. Fintechs and banks now have to assume customers will route basic money questions through general assistants—API strategy and brand surface area matter more than app stickiness.