Anthropic Taught Claude to Fill Carts. It Skipped the Register on Purpose.

On September 2, 2026, Anthropic gave any retailer free code to run a shopping agent and a merchant agent on Claude, and it left out the one piece most people expected: payment.

The release, called the Commerce Agents blueprint, is open source on GitHub and ships with a plugin for Claude Code. An engineering team can wire it into an existing storefront in days instead of months.

Anthropic named Shopify, Priceline, Visa, Mastercard and Accenture as early partners, timed nine weeks ahead of the U.S. holiday shopping season. For a small commerce or retail-tech team, the real question isn't whether Claude can do this job. It's whether Anthropic's own numbers are the ones to trust before spending a sprint on it.

What Anthropic actually shipped, and why Claude now has three jobs

The blueprint splits into two roles. A shopper agent searches a catalog, compares products, builds a cart inside the conversation and hands it off to checkout; it also answers order-status, return and refund questions.

A merchant agent works the back office. It flags inventory problems, recommends discounts from sales history and drafts marketing campaigns, though Anthropic says operational changes still need a human's approval before they go live. Reference implementations ship for retail, travel, telecom and ticketing, and the code deploys through the Claude API, Amazon Bedrock, Microsoft Foundry or Google Cloud Vertex AI, according to Anthropic's own announcement.

Anthropic's own announcement, captured 4 September 2026. The 35%/60% figures shown here are the company's own, discussed further below.

That announcement makes plain what buying Claude actually means in 2026, and it isn't one decision. Anthropic charges $17 to $20 a month for the chat app, up to $125 for a premium team seat, according to its current pricing page.

Claude Code rides free on top of any paid plan, drawing from the same usage pool. Heavier teams bill it separately through pay-as-you-go API credits instead. The commerce blueprint has no price tag at all. It's a GitHub repository, not a subscription tier, and whatever it costs a retailer shows up as cloud usage, not a Claude invoice. Three different products, three different ways of paying for them, one brand name over all of it.

Claude's pricing page, captured 4 September 2026. The commerce blueprint doesn't appear here because it isn't a plan; it's free code that bills as cloud usage instead.

Who this actually helps

An engineering team that already wanted to prototype a shopping agent gets a running start. The harnesses, guardrails and prompt patterns that took the reference implementations months to build now ship as a starting point instead of a research project. That costs nothing to try.

It also means Anthropic is repeating a decision OpenAI made the hard way. OpenAI's own Instant Checkout let ChatGPT complete a purchase directly inside the chat window, starting in September 2025.

By March 2026, OpenAI repositioned it. A purchase now routes through the merchant's own app instead of resolving inside ChatGPT, a change the company tied to how often tax, inventory and pricing data need updates that only a merchant's own system keeps current. Anthropic's blueprint never puts Claude between the shopper and the checkout page at all. It hands the pattern to the retailer's engineers and gets out of the way.

If your team isn't sure yet whether it needs an agent this specific, or something closer to a general recommendation engine, that's worth running through Smart Match before it's a question about which blueprint to fork.

Who this hurts

The shopper agent's support function, answering questions about an order, a return or a refund, is the entire product that Parahelp, Pylon and Sierra already sell, priced per resolution or per seat. Folding that function into a free code template from the model vendor itself is a direct threat to point solutions built on exactly that job.

It's a familiar squeeze. The same thing hit customer-support vendors once when general-purpose assistants started answering the same tickets without a dedicated contract.

The merchant agent's marketing-draft feature lands on the same ground Revnu already occupies: an autonomous system built specifically to generate and run marketing campaigns for a growth team. Revnu still does more around that one job than a blueprint's single feature does. But "more than a checkbox in someone else's template" is a narrower moat than it was on September 1.

The turn: the numbers are Anthropic's own

Anthropic's headline claim is that retailers running shopping agents on Claude have seen carts grow as much as 35% larger, with shoppers 60% more likely to complete a purchase. No retailer is named next to that number, and no outside auditor checked it.

PYMNTS reported plainly that these are "the company's own reported figures rather than an independently verified benchmark." That framing matters more here than in most vendor announcements, because the number is the entire pitch to a retailer deciding whether an engineering sprint is worth it.

The skepticism has company. A PYMNTS Intelligence survey found 43% of retailers are already piloting autonomous AI shopping tools. A separate PYMNTS Intelligence figure put consumer unease at 95%, the share of shoppers who report at least one concern about agentic commerce, with only 5% reporting none.

Retailers are moving faster than their own customers are comfortable with. Anthropic's blueprint doesn't touch that gap. It ships no payment protocol, no product catalog and no advertising layer of its own, leaving the parts shoppers actually distrust to the partners, Visa and Mastercard among them, that already carry that trust.

What to watch

Watch whether any of the five named partners publishes its own measured before-and-after numbers before Black Friday, November 27, 2026. If none does, the 35% and 60% figures will still belong to Anthropic alone eight weeks after the announcement, at the exact moment retailers most need proof that isn't self-reported.

Anthropic solved the engineering problem in one release. Whether it solved the trust problem depends on whether a single retailer says these numbers out loud before the last shipping deadline of December.

Frequently asked questions

What is Claude actually used for in 2026?

Claude is really three separate products under one brand: a consumer chat app for writing and research, Claude Code for software engineering work, and, as of September 2026, an open-source blueprint that lets retailers build shopping and merchant agents on the same models. Each has its own pricing and its own buying decision.

What are Anthropic's new Commerce Agents for Claude?

Released September 2, 2026, the Commerce Agents blueprint is open-source code for two agent types: a shopper agent that searches a catalog, builds a cart and answers order questions, and a merchant agent that flags inventory issues, recommends pricing and drafts marketing campaigns. It ships with a Claude Code plugin and deploys through the Claude API, Amazon Bedrock, Microsoft Foundry or Google Cloud Vertex AI.

Does Claude handle payment or checkout for retailers?

No. Anthropic's blueprint explicitly leaves out a payment protocol, a product catalog and an advertising layer, handing the cart off to the retailer's existing checkout instead. That mirrors a lesson OpenAI already learned: its own Instant Checkout tried resolving purchases directly inside ChatGPT starting in September 2025, then by March 2026 rerouted purchases through each merchant's own app instead.

Is Claude Code included with a Claude subscription?

Yes. Claude Code is bundled at no extra charge with any paid Claude plan, from the $17-20 monthly Pro tier up to Team and Enterprise seats, and draws from the same usage allocation as chat. Teams with heavier coding workloads can instead buy pay-as-you-go API credits through a Console account.

Are Anthropic's commerce agent results independently verified?

No. Anthropic's claim that shopping-agent carts grow up to 35% larger, with shoppers 60% more likely to complete a purchase, comes with no named retailer and no outside audit attached. PYMNTS reported these are the company's own figures rather than an independently verified benchmark, and separate PYMNTS Intelligence research found 95% of consumers report at least one concern about agentic commerce.

Covered in this guide

• Claude: Claude API with Sonnet 5 (agentic, near-Opus performance at half the cost) and Claude Science (auditable research workbench). Free to Pro.
• OpenAI: ChatGPT is OpenAI's AI assistant with 900 million weekly users and GPT-5.6 Sol, covering writing, coding, image generation, and web search with a free plan and Plus at $20/month.
• Anthropic: Anthropic, founded 2021 by 7 ex-OpenAI researchers, builds Claude and was valued near $965B after its May 2026 Series H round.
• Claude Code: Claude Code by Anthropic scores 88.6% on SWE-bench Verified, starts at $20/month with no free tier. Reads 1M-token codebases, edits files, runs tests, and opens PRs across 8 platforms.
• Parahelp: AI support agent from an 11-person YC-backed team that resolves complex tickets end-to-end via email, chat, and Slack for Perplexity, Framer, and Replit.
• Pylon: Pylon is a B2B customer support platform where AI agents pre-investigate every ticket across Slack, email, and chat; trusted by 1,500+ companies.
• Revnu: Y Combinator-backed AI growth platform that runs SEO, paid ads on 4 channels, outbound outreach, and A/B testing autonomously for software startups.
• Sierra: Enterprise AI agent platform from Bret Taylor valued at $15.8B in 2026, used by 40% of the Fortune 50 for customer service across chat, voice, and email.

Sources