{{current_date_full_with_abbr_day}}
3-minute read
3
AI UPDATES
2
DEMOS
1
READ
👋 Good morning.
Kroger expected its AI shopping assistant to shrink baskets. It grew them: customers freed from meal planning spend the time they saved adding things to the cart, per Supermarket News. Efficiency, redirected.
IN 3 MINUTES, GET 1% BETTER THAN EVERYONE YOU KNOW:
Run real work by voice on your phone
Gemini plugs into Linear, Airtable and 12 more
Claude opens a shop for tools and agents
$25 of Opus built a physics lab in 86 minutes
What MCP is, in plain English
3 AI updates
1. 🎤 Talk, then work
OpenAI put voice into ChatGPT's Work tab on mobile on Wednesday, per TechCrunch. Plus and Pro can now say "draft that email" or "summarise the Slack thread" and get it done on the phone, along with building sites, making presentations and running the cloud browser. Free and Go get plugins and connected apps. You can swap between talking and typing inside one conversation, and pick it up on desktop later.
Do this: On tomorrow's commute, talk one real email into the Work tab and finish it at your desk.
Hands free. Inbox not so much.
Would you run real work by voice on your phone?
2. 🔗 One chat, more apps
Google connected 14 more apps to Gemini on Wednesday. For work: Airtable, Linear, monday.com, PandaDoc, Wispr AI and Zoho. For design: Adobe, Picsart, Squarespace and Webflow. For everything else: apartments.com, Experian, Peloton and SeatGeek. Turn them on in Gemini's settings, then pull one into a chat by typing @ or just naming it. Ask what's blocked in Linear without opening Linear.
Do this: Connect the one app you check most, then ask Gemini a question you'd normally open a tab for.
Fewer tabs. Same deadlines.
3. 🛍 Marketplace, stocked
Claude Marketplace opened to browsing on Wednesday, per Anthropic. Three shelves: connectors and plugins like Slack and Notion, agents and products you can buy from companies like Cursor and CrowdStrike, and service partners like Accenture and Deloitte if you need people to run it. Earlier this month the marketplace was a way to spend an existing Anthropic contract. Now it's a place to look around.
Do this: Add the one connector you keep alt-tabbing to, Slack or Notion, before you write another glue script.
Shelf stocked. Cart still optional.
2 Things you didn't know you can do with AI
1. 📷 Ask for a lab, not an explanation · Opus 5.5
Ryan Sael asked Opus 5.5 to explain camera focus by building an interactive lens lab. It came back in one shot: 1 hour 26 minutes, $25.66 of API. Turn the focus ring at lens.lab.sael.net and watch the sharp plane sweep the scene. Steal the prompt shape for anything you can't picture. The catch: one shot cost $25, so ask for one thing, not a course.
2. 🍉 A short film about a watermelon · Opus 5.5
Anthropic is running a thread of early Opus 5.5 experiments. This one, by Kevin Ngo, is a 29-second story about a watermelon, coded rather than filmed. It's the clearest look at what the model does when you ask for a story instead of a spreadsheet. The catch: pure inspiration. Nothing to install.
Follow the thread to watch many more examples.
1 Read
Our MCP explainer unpacks the acronym you keep seeing in every tool announcement, including three in today's issue. Model Context Protocol is an open standard for letting AI models talk to your tools and data. Think USB-C: one cable instead of a different adapter for every device. OpenAI, Google, Microsoft, Cursor and Windsurf have all adopted it, there are over 5,800 servers, and you can start in Claude Desktop without writing code.
The catch: a badly configured MCP server is a real attack surface, and the piece names the documented ones. Read that part before you connect anything that touches customer data.
One cable. Fewer custom bridges.
Don't forget to rate today's post
That's your 1% for today. Reply with the first thing you'd dictate on the way to work.
Kushank, AI teammates, and the human team
PracticalyAI
