👋 Hey, Linas here! Every day, I break down 3 stories shaping the future of FinTech & Artificial Intelligence - plus the money movements and trends worth tracking. First time here? 400k+ FinTech and AI leaders get this daily. Join them:
On September 2, 2026, Anthropic released Claude Commerce Agents, an open blueprint for building AI shopping agents and merchant-operations agents on Claude. The rest of the industry is still fighting over the consumer gateway: Google and Shopify run live agentic checkout in AI Mode and Gemini through the Universal Commerce Protocol, Microsoft’s Copilot Checkout counts more than 500,000 participating merchants, Perplexity sells in-chat through Instant Buy with PayPal, and Amazon’s Buy for Me completes purchases from inside its app. OpenAI tried the same play and failed. It retired ChatGPT’s in-chat Instant Checkout in March 2026, roughly five months after launch, after flagship partner Walmart found conversion inside the chatbot ran about three times below its own site. ChatGPT now hands shoppers to retailer apps and merchant sites to complete the purchase.
That retreat revealed exactly the world Anthropic built for: shoppers discover in AI but buy on the merchant’s property. Claude Commerce Agents is therefore working agent code that lives inside the merchant’s own website, app, and back office.
The release contains shopping- and merchant-agent implementations, four runnable vertical examples, a safety harness, eval patterns, a Claude Code builder plugin, and deployment paths across the Claude API, Amazon Bedrock, Microsoft Foundry, and Google Cloud Vertex AI. It is not a consumer shopping destination, a commerce protocol, or a payment processor: nothing in the reference implementation places an order or charges a card. The narrow scope is the whole strategy. Anthropic wants Claude to be the intelligence and runtime inside merchants’ experiences while catalogs, checkout, payment, fulfillment, customer ownership, and regulatory responsibility stay with merchants and their infrastructure providers.
The launch matters for five key reasons:
It moves Anthropic from model vendor toward application architecture owner. The blueprint prescribes how a commerce agent should be built, rather than only which model it should call, and it runs the same code across Anthropic’s API and all three major clouds.
It is a merchant-control counter-position to AI shopping gateways. While rivals intermediate the merchant, Anthropic starts with an agent embedded in the merchant’s own property and an agent used by the merchant’s staff. Payments remain the merchant’s choice, and checkout hands off to the host.
It extends a repeatable verticalization factory. Claude for Financial Services, finance-agent templates, Claude for Small Business, and now Commerce Agents package the same ingredients in different proportions: frontier models, Skills, connectors, controlled actions, and Managed Agents. Anthropic is simply standardizing the horizontal stack and using vertical blueprints to create demand for it.
It responds to a distribution asymmetry. As per Similarweb, Claude AI reached 107.6M unique web visitors in July 2026, up from 25.3M in January, but still well below ChatGPT’s 469.1M and Gemini’s 302.2M. Merchant embedding hence lets Anthropic reach shoppers through other companies’ distribution.
It turns Anthropic’s safety positioning into executable commerce controls. Server-issued IDs, transaction caps, staged changes, approval gates, content sanitization, and snapshot evals sit in code rather than only in prompts. The model cannot charge a card or apply a price change in the reference implementation.
At the core, it’s all about the control of the layer that translates commercial intent into calls across catalog, customer, inventory, pricing, campaign, order, and payment systems - better chat on a store is incidental. If Anthropic can make Claude the default reasoning engine and Managed Agents the default runtime for that layer, commerce becomes a high-frequency, measurable source of model demand. But if open protocols (i.e., GLM-5.3 Flash or Qwen3.8-27B) and model routing make intelligence interchangeable, most of the value will stay with Shopify, Salesforce, payment networks, payment service providers (PSPs), marketplaces, and the merchants that own demand, identity, transaction data, and systems of record.
If you’re a founder, builder, operator, or investor working in or exploring the agentic economy, this deep dive is for you. Inside:
What actually shipped and where the hard boundaries sit
The recommended architecture and what a commerce agent really costs to run, including the cost math on Sonnet 5 versus Opus 5
What the early evidence proves and what it doesn’t: the 35% cart claim, Adobe’s conversion data, and Anthropic’s own agent-negotiation experiments
Why OpenAI’s checkout retreat is the strongest validation yet of the merchant-side model
The full competitive map across OpenAI, Google, Salesforce, Shopify, Microsoft, Amazon, Stripe, and Visa
What merchants, commerce platforms, payments players, and investors should each do next
Anthropic’s monetization paths, the bear case, and the production failure modes to watch
The scenarios and indicators that will decide who captures the most value through 2031
The architecture decisions merchants lock in over the next two quarters, ahead of the first true agentic holiday season, will be expensive to unwind and will shape who owns the customer, the data, and the margin for years. Everything below comes from the repository code, Anthropic's engineering guide, the published pricing, and what actually happened to each competing checkout program.
It should give you enough to decide where to build, partner, or invest before the market decides for you.