Good morning, AI enthusiasts. AI shopping was headed for chaos, every platform building its own checkout, every agent speaking a different language. Google and Shopify just stopped that from happening.
The Universal Commerce Protocol creates one standard for AI agents to shop anywhere, with Target, Walmart, and 20+ retailers already on board. Is this the infrastructure that keeps AI commerce from breaking into walled gardens?
In today's recap:
Google and Shopify's universal shopping standard for AI agents
Anthropic's Cowork brings Claude Code to non-developers
Turn Gemini into your personal AI tutor
Ramp shares blueprint for its coding agent Inspect
5 new AI tools and more
GOOGLE & SHOPIFY
Google and Shopify launch open AI shopping standard
Source: Google
Recaply: Google and Shopify just launched Universal Commerce Protocol. It's an open standard that lets AI agents buy things across any platform with over 20 major retailers signed on, including Target, Walmart, Etsy, and Wayfair.
Key details:
The protocol handles the full shopping journey from discovery to support after purchase. It works with existing standards like Agent2Agent, Agent Payments Protocol, and Model Context Protocol.
Major retailers backing UCP include Shopify, Etsy, Wayfair, Target, Walmart, Best Buy, Macy's, The Home Depot, and Flipkart. Payment partners include Adyen, American Express, Mastercard, Stripe, and Visa.
Shopify merchants can now sell directly in AI Mode in Google Search and the Gemini app through embedded checkout. It supports Google Pay and PayPal, with retailers keeping control as the seller of record.
The feature rolls out soon for eligible US retailers, with global expansion and additional capabilities like loyalty rewards and custom shopping experiences coming in the following months.
Why it matters: Google and Shopify built the first major open standard for agentic commerce, betting that AI shopping won't be controlled by one platform. With Target, Walmart, and payment giants on board, UCP stops AI shopping from breaking into competing systems. The protocol is flexible. Retailers can adjust checkout flows for discount codes and subscriptions. This matters because millions of shopping talks happen daily in ChatGPT, Gemini, and Claude. If UCP becomes the standard, merchants set up once and sell everywhere AI conversations happen.
ANTHROPIC
Anthropic launches Cowork for non-coding work
Source: Anthropic
Recaply: Anthropic just introduced Cowork in research preview. The tool brings Claude Code's capabilities to anyone through a simpler interface after developers began using Claude Code for tasks far beyond coding.
Key details:
Cowork lets users give Claude access to a chosen folder where it can read, edit, or create files while handling multi-step tasks on its own. It can reorganize downloads or create spreadsheets from screenshots.
The tool works with existing connectors and has built-in skills for documents and presentations. It connects with Claude in Chrome for browser tasks. Users can queue multiple tasks at once.
Cowork is built on the Claude Agent SDK, the same foundation as Claude Code. It asks before taking major actions. Users need to give clear instructions since Claude can delete files if told to do so.
Max subscribers can access Cowork now through the macOS app. Cross-device sync and Windows support are coming. Anthropic plans safety improvements based on user feedback.
Why it matters: Anthropic noticed developers were using its coding agent for everything. They organized media files and scanned social posts. This showed a broader need for AI that works on its own, not just for code. Cowork opens this up without requiring command-line skills or virtual environments. Product managers, designers, and anyone else can now use Claude's full capabilities. The challenge is giving Claude more control while keeping users safe. Anthropic uses folder permissions and confirmation prompts to help. Prompt injection risks still exist across the industry.
RAMP
Ramp releases blueprint for Inspect coding agent
Source: Ramp
Recaply: Fintech company Ramp just released the blueprint for Inspect, its in-house coding agent that writes 30% of merged pull requests across frontend and backend repos after reaching that adoption level in just months.
Key details:
Inspect runs in sandboxed cloud environments on Modal with full dev setup including Vite, Postgres, and Temporal. It connects to Sentry, Datadog, LaunchDarkly, GitHub, Slack, and Buildkite for complete context like a Ramp engineer would have.
Sessions start near instantly using file system snapshots built every 30 minutes. Builders can launch unlimited sessions at once. No need to manage local resources or keep laptops running.
The system works across Slack, web, and a Chrome extension with visual editing. It includes voice input. All changes sync across interfaces. Multiple team members can work in one session together.
Ramp built Inspect using Modal for sandboxes and Cloudflare Durable Objects for the API. Each session gets its own SQLite database. WebSocket streaming runs on Cloudflare's Agents SDK.
Why it matters: Ramp hit 30% adoption in months without forcing anyone to use it. This proves that building your own tooling beats buying off-the-shelf agents for internal work. The key insight is verification. Inspect can run tests, check telemetry, and query feature flags. For frontend work, it takes screenshots to verify changes visually. It has all the context to prove its work like an engineer would. By sharing the blueprint, Ramp is betting that fast, free, unlimited coding sessions will become expected. Anyone can build this if they know the architecture.
NEWS
Beyond the big stories in AI today
OpenAI acquired Torch for $60M, a healthcare tech startup building a "unified medical memory" that aggregates patient data across providers to enable longitudinal health tracking.
Anthropic announced Claude for Healthcare, a HIPAA-ready version with new tools that connect to medical databases like CMS and ICD-10 to help doctors access health data faster.
Apple and Google announced a multi-year partnership to use Google's Gemini AI in Siri, giving the voice assistant better understanding of context and more natural conversations.
Owkin released AI agents for biology research that can automate lab work, plus a Pathology Explorer tool that works with Claude to analyze medical images.
Meta launched Meta Compute, a plan to build massive AI data centers with tens of gigawatts of power to handle the company's growing AI projects.
NVIDIA and Eli Lilly announced a $1B partnership over five years to create a research lab using NVIDIA's Vera Rubin AI chips for drug discovery.
GUIDES
Turn Gemini into your personal AI tutor
Recaply: In this tutorial, you will learn how to use Gemini's Guided Learning mode to get step-by-step tutoring that breaks down problems instead of just giving you answers, helping you actually understand concepts at your own pace.
Step-by-step:
Go to gemini.google.com or open the Gemini app, click the Tools selector at the bottom of the chat, then select "Guided Learning" from the dropdown menu
Describe what you want to learn with context about your level, like "Teach me Python basics as a beginner" or "Help me understand how blockchain works for a business presentation"
Answer the questions Gemini asks about your current knowledge level and learning goals to personalize the tutoring approach
Work through the step-by-step guidance as Gemini breaks down concepts with questions, images, diagrams, and interactive quizzes instead of direct answers
Ask follow-ups like "Can you explain that differently" or "Give me a practice problem" to deepen understanding, then switch back to regular mode when you need quick facts
Pro tip: Connect your Google Drive in settings so Gemini can reference your class notes, work documents, or study materials for tutoring that's tailored to exactly what you're learning.
TOOLS
Trending AI Tools
💻 Cowork - Anthropic's tool brings Claude Code's capabilities to anyone through a simpler interface
🔧 Universal Commerce Protocol - Open standard that lets AI agents buy things across any platform
🎨 Midjourney Niji V7 - Image model made for anime
🎤 Scribe v2 - ElevenLabs' speech-to-text model that writes down what people say
⚙️ NousCoder-14B - AI coding model trained to solve hard programming problems
