This Week in Agentic (W13)

Weekly recaps—journalistically speaking, pretty much the most stuffy thing you can imagine. But, friends of the sun, there’s so much going on out there at such a breakneck pace that an overview isn’t such a bad idea after all.

Shopify: Millions of stores, discoverable in ChatGPT

Shopify has launched Agentic Storefronts. Since March 24, Shopify stores (provided they’re based in the U.S. and sell to a U.S. audience) can be found automatically in ChatGPT—no setup, no app installation required. The key point: Checkout takes place on the merchant’s storefront, not in ChatGPT. An in-app browser overlays the chat, and Shop Pay handles the transaction. Subscriptions, bundles, pricing—everything stays with the merchant.

OpenAI: Shopping pivot, Sora shelved

In parallel with the checkout rollback, OpenAI launched Visual Shopping in ChatGPT—product discovery instead of transactions. The new narrative: We are a discovery platform, not a payment processor. And then OpenAI shut down Sora—the video platform that had launched just months ago. (Funny, didn’t Disney want to invest $1 billion in OpenAI and license 200+ characters for Sora?) Focus instead on enterprise and coding tools. More like Anthropic, essentially.

Anthropic Launch Watch

Speaking of which: Anthropic is currently rolling out features on a weekly basis. This week:

• Computer Use is coming to Claude Code and Cowork. Claude can now operate its own computer—open browsers, fill out forms, switch between apps. The step from coding assistant to desktop agent. (And the ultimate reason to consider extra sandbox hardware.)
• Claude Code Channels: Persistent coding agents, controllable via Telegram and Discord. Send a task, the agent works asynchronously, and the result comes via message. Admin controls and sender whitelists address the security concerns raised by open-source alternatives like OpenClaw. In other words: Developers can now code on the go, in bed, and in the bathroom.
• Safer Auto Mode: Claude Code gets a tiered autonomy mode—more freedom for non-critical actions, confirmation for risky ones. The fundamental problem with any agent architecture: how much leeway do you give the system?

Also:

• Mistral’s CEO is calling for a content tax on AI companies in Europe.
• OpenAI is offering PE firms a 17.5% guaranteed return—sounds like desperation, not strength.

An LLM researched and wrote. A human read, edited, and approved it. We’re still debating which of the two did more work.