- unwind ai
- Posts
- Vibe Work with Claude Cowork
Vibe Work with Claude Cowork
+ Google’s open-source Universal Commerce Protocol for AI shopping agents
Shubham Saboo & Gargi Gupta
January 13, 2026
In partnership with
Today’s top AI Highlights:
& so much more!
Read time: 3 mins
AI Tutorial
Evaluating startup investments requires hours of research across multiple domains - company analysis, market research, financial modeling, and risk assessment. This setup literally automates this entire workflow with AI agents that work together like a real investment team.
In this tutorial, you'll build an AI Due Diligence Agent Team using Google's Agent Development Kit (ADK) and Gemini 3 models, and Nano Banana.
This 7-agent team researches any startup (from early-stage unknowns to well-funded companies), analyzes the market, builds financial projections, assesses risks, and generates professional investment reports - all autonomously with seamless handoffs and a sophisticated analysis with reports.
We share hands-on tutorials like this every week, designed to help you stay ahead in the world of AI. If you're serious about leveling up your AI skills and staying ahead of the curve, subscribe now and be the first to access our latest tutorials.
Latest Developments
When developers got their hands on Claude Code, something unexpected happened. They used it for coding, sure, but then they kept going - reorganizing downloads, building spreadsheets from screenshots, converting scattered notes into polished reports.
Anthropic watched this unfold and realized people wanted an AI that actually works alongside them, not just answers questions.
Enter Claude Cowork, a research preview that makes agentic Claude accessible to anyone on macOS with a Claude Max subscription. Point it at a folder on your machine, and Claude handles the rest, like reading files, making edits, creating new documents, all while keeping you informed about its progress.
You're constantly feeding context and reformatting outputs, you’re delegating your work to a coworker. You can even pair it with Claude's existing connectors and new document creation skills to handle tasks that span multiple tools and formats.
Key Highlights:
Real delegation, not conversation - Queue multiple tasks and let Claude work through them in parallel without waiting for one to finish before starting another.
Autonomous task execution - Claude creates a plan and works through it independently, asking for approval before significant actions while you queue up additional tasks in parallel.
Folder-based access control - You choose exactly which folders and connectors Claude can see and interact with, maintaining full control over what gets accessed or modified.
Browser tasks with Skills - Just like Claude Code and claude.ai, Skills improve Claude's ability to create documents and presentations, and Chrome integration automates the browser-based tasks.
Safe early access - Research preview includes prompt injection defenses and clear guidance on secure usage, with cross-device sync and Windows support coming soon. Join the waitlist if you're not on Max plan.
Vibe code with your voice
Vibe code by voice. Wispr Flow lets you dictate prompts, PRDs, bug reproductions, and code review notes directly in Cursor, Warp, or your editor of choice. Speak instructions and Flow will auto-tag file names, preserve variable names and inline identifiers, and format lists and steps for immediate pasting into GitHub, Jira, or Docs. That means less retyping, fewer copy and paste errors, and faster triage. Use voice to dictate prompts and directions inside Cursor or Warp and get developer-ready text with file name recognition and variable recognition built in. For deeper context and examples, see our Vibe Coding article on wisprflow.ai. Try Wispr Flow for engineers.
AI shopping agents can now talk to any retailer's checkout system in the same language.
Google just launched the Universal Commerce Protocol (UCP), an open standard that lets AI agents handle the entire purchase journey without needing custom integrations for every platform or business.
Co-developed with Shopify, Etsy, Wayfair, Target, and Walmart, UCP works as a universal translator between consumer surfaces, businesses, and payment providers. It's compatible with existing protocols like MCP, A2A, and Agent Payments Protocol (AP2), meaning merchants can plug into multiple AI platforms without rebuilding their infrastructure.
Starting soon, you'll see native checkout powered by UCP directly in Google's AI Mode and Gemini app, where shoppers can complete purchases using Google Pay or PayPal without leaving the conversation.
Key Highlights:
Modular architecture - Merchants pick specific capabilities and extensions they want to support, from basic checkout to loyalty programs and discount codes. The protocol handles everything through standard retail operations that map directly to existing business logic.
Multiple integration paths - Supports REST APIs, MCP binding, Agent2Agent (A2A), and works with existing payment handlers. This means you can choose the communication method that fits your development stack without rewriting core systems.
Merchant-controlled experience - Retailers remain the merchant of record and own all customer data and relationships. They can customize checkout flows, apply member pricing, offer loyalty enrollment, or suggest product add-ons while maintaining full control over the transaction.
Production-ready starting point - Google Merchant Center already has millions of SKUs and product catalogs in place, giving merchants with existing feeds a fast path to agentic commerce without building from scratch.
Quick Bites
230M people ask ChatGPT health questions. Now it's official.
OpenAI launched ChatGPT Health, a sandboxed space where users can connect medical records and wellness apps (Apple Health, MyFitnessPal, etc.) for personalized health conversations. The feature, built with input from 260+ physicians, lives separately from regular ChatGPT, uses dedicated encryption, and won't train on your health data. Currently rolling out to early users on the waitlist, with broader access coming in weeks.
AI Agent Memory is done; self-learning is here
Agno, the open-source AI agent framework, just shipped Learning Machines, a framework that reframes agent memory as an active learning process. The system uses a simple five-function protocol that lets you build custom "learning stores" in about 50 lines of code, whether for user profiles, legal docs, or sales pipelines, with built-in modes ranging from invisible background learning to human-approved updates. The real hook is that agents can eventually analyze their own failures and propose behavioral changes without retraining. The architecture supports everything from automatic background learning to human-in-the-loop approval. Check out the PR here.
Just like Search, AI Overviews is coming to Gmail
Gmail is getting Search-like AI Overviews, powered by Gemini 3. Ask your inbox questions in natural language and get instant answers instead of keyword hunting through threads. The feature synthesizes entire conversations into summaries and can surface specific details like "who quoted me for that bathroom renovation last year?" without the archaeological dig through your messages.
Another big news is that Help Me Write and Suggested Replies in Gmail are rolling out to everyone at no cost.
Qwen's new open-source multimodal RAG stack
Qwen just dropped their multimodal embedding suite - Qwen3-VL-Embedding and Qwen3-VL-Reranker, built on their Qwen3-VL foundation model. The two-stage pipeline handles text, images, screenshots, and video inputs, with the 8B embedding model scoring 77.8 on MMEB-V2 (currently first place). Both models come in 2B and 8B sizes, support 30+ languages, and let you customize task-specific instructions for that 1-5% performance bump. Available to download on Hugging Face and Modelscope.
Anthropic blocked third-party use of Claude Code subscriptions
Anthropic (temporarily) blocked third-party usage of the Claude Code subscriptions. Clearly, the $200/month all-you-can-eat plan delivers $1,000+ worth of API tokens, but there’s a big chunk of developers using other agent harnesses like Opencode, connected with their Claude subscriptions. The company mentioned Terms of Service violations and telemetry issues, and received a LOT of backlash from developers. For now, they've lifted those bans and promised clearer OAuth warnings.
Tools of the Trade
Flux - Create custom AI agents that run natively in iMessage - you describe what you want, and in two minutes you've got a chatbot that anyone can text without downloading an app. The premise is simple: people actually open their messages constantly, unlike the AI apps collecting dust on their home screens.
Mantic - A code search engine that ranks files in < 500ms using path patterns and metadata instead of reading content or running embeddings. Built as an MCP server, it cuts token waste by filtering files deterministically before any content gets loaded into context.
Claude-reflect - A Claude Code plugin that automatically captures your corrections and preferences during coding sessions, then lets you review and sync them to CLAUDE.md files so Claude remembers them permanently. It uses hooks to detect patterns like "no, use X instead of Y" or "remember: always do Z," queues them with confidence scores, then prompts you to approve additions.
Awesome LLM Apps - A curated collection of LLM apps with RAG, AI Agents, multi-agent teams, MCP, voice agents, and more. The apps use models from OpenAI, Anthropic, Google, and open-source models like DeepSeek, Qwen, and Llama that you can run locally on your computer.
(Now accepting GitHub sponsorships)
Hot Takes
Some of the best software engineers are losing their jobs to AI
everyone I know in tech is either depressed or euphoric right now, there is no middle ground
That’s all for today! See you tomorrow with more such AI-filled content.
Don’t forget to share this newsletter on your social channels and tag Unwind AI to support us!
PS: We curate this AI newsletter every day for FREE, your support is what keeps us going. If you find value in what you read, share it with at least one, two (or 20) of your friends 😉
Reply