Today in AI — 26 March 2026

AI agents got wallets today. The commerce stack is being rebuilt for a world where the buyer might not be a person, and vertical AI companies are raising the capital to put agents into law firms, hospitals, and retail back offices. The question is no longer whether agents will act autonomously but who sets the rules when they do.

Agents learn to shop

OpenAI tried owning the checkout flow and failed. Low conversion rates killed Instant Checkout, so now ChatGPT becomes a discovery layer that hands off to retailer-built apps. Gap went the opposite direction with Google, embedding direct checkout inside Gemini using the Universal Commerce Protocol. Shopify unified both approaches: Agentic Storefronts let any brand sell across ChatGPT, Gemini, and Copilot from one admin panel. And Stripe's Machine Payments Protocol takes it further, giving agents the ability to spend money on their own. Four different architectures for agentic commerce, all shipping in the same week.

• OpenAI kills Instant Checkout and pivots ChatGPT shopping to visual discovery and retailer apps — CNBC
• Gap becomes the first major fashion retailer to offer direct checkout inside Google Gemini — CNBC
• Shopify opens Agentic Storefronts and the Universal Commerce Protocol to every brand — Shopify
• Stripe's Machine Payments Protocol lets AI agents spend money without human intervention — Stripe

Vertical AI raises big

Harvey at $11B, Granola at $1.5B, Qualified Health pulling in $125M, Glimpse grabbing $35M, Moda at $7.5M. The pattern: domain-specific AI companies that embed agents into an industry's actual workflow (filing legal briefs, disputing retail deductions, managing hospital revenue) are attracting serious capital. Granola's pivot from meeting notes to "enterprise context layer" is the most telling move. The value isn't in the transcript, it's in making organisational knowledge available to every agent in the stack.

• Harvey raises $200M at $11B valuation to scale AI agents across law firms — CNBC
• Granola raises $125M at $1.5B valuation as it pivots from meeting notes to enterprise context layer — TechCrunch
• Qualified Health raises $125M to become the AI operating system for US hospitals — STAT News
• Glimpse raises $35M from a16z to automate the $50B retail deductions problem with AI agents — TechCrunch
• Moda raises $7.5M to give every professional an AI design agent that learns your brand — GlobeNewsWire

Trust fractures

WebinarTV is scraping public Zoom links and turning meetings into podcasts without consent. A sponsored Google ad for "Install Claude Code" was serving malware from a fake Anthropic page. As agents get more capable, the attack surface and the privacy violations grow with them.

• 404 Media reveals WebinarTV is secretly recording Zoom calls and turning them into AI podcasts — 404 Media
• Hackers plant malicious Claude Code install page as top Google search result — 404 Media

Platform plays

OpenAI's Foundation is pledging $1B for AI safety research and workforce impact. Google is building a memory-transfer tool to make switching from ChatGPT to Gemini frictionless. Both are playing the long game: OpenAI by investing in legitimacy, Google by lowering switching costs.

• OpenAI Foundation pledges $1B to tackle AI's impact on jobs, biosecurity, and children — Fortune
• Google builds a ChatGPT-to-Gemini migration tool to poach rival users with memory transfer — 9to5Google

Frontier models hit a milestone

Four frontier models independently solved an open Ramsey theory conjecture, the first time AI has cleared FrontierMath's open-problem track. When multiple models converge on the same result, it points to a new capability regime rather than a one-off trick.

• Frontier AI models solve an open Ramsey theory conjecture for the first time — Epoch AI

If you're building products that touch commerce, the protocol wars (Universal Commerce Protocol, Machine Payments Protocol, Agent Sizing Protocol) matter more right now than the model wars. Pick the rails carefully.

---