Tuesdays are 3 AI updates, one hot take, one action. Fridays are 1 practical real estate agent workflow.

All three stories this week connect to the concept of how much information do you want an assistant to see? The recent Black October moment means that privacy is gonna be more of a focus for all of the AI companies over the next few months. Here are three new updates that show you different ways information is being shared across the new ecosystem and how they’re important for agents.

Anthropic launched Claude Duet, an assistant that checks your connected apps and tells you what needs attention before you ask. Announced October 21, it sends a short morning brief and looks through your apps every few hours, working in its own cloud browser, and it asks for your approval before it sends a message, pays for something, signs a document or deletes a file. It's a research preview for US Pro and Max subscribers, by waitlist, on the web and in a phone app, and its background checks count against your normal plan limits. Meta and OpenAI have shipped similar always-on assistants, so I'd expect most agents to have one watching their inbox within a year.

Zillow put its home search inside Claude, so a buyer can ask for homes there and never have to go to Zillow. According to Zillow's October 19 release, anyone who adds the connector under Customize, then Connectors, can ask for three-bedroom homes in Denver under $600,000 and get listings, maps, value estimates and affordability math in the conversation. Each listing keeps the broker and MLS credit it carries on Zillow.com, and the buyer can request a tour or ask to be connected with an agent without leaving the chat. Zillow says it sends no data to Anthropic, but the release doesn't say how agent requests get routed. Add it yourself this week and search your own listing the way a buyer would, because that's the screen your client may see first.

ChatGPT now lets you lock a conversation behind a PIN, which hides it from your history, pinned chats and search. In the app or on the web you open a chat's menu, tap the three dots, choose Lock chat, and reopen it later with a four-digit PIN or Face ID. OpenAI's release notes say it started rolling out October 20 and is gradual, so you may not see it yet. A few notes about the limits: there's no added encryption, your data and model-improvement settings stay as they were, memories saved before you lock a chat remain, and resetting a forgotten PIN permanently deletes every locked chat. I could see a use case where you use it when you have a client sitting next to you while you share your screen, but I wouldn't rely on it too much at this point as this is the first experiment with this kind of privacy control.

Open ChatGPT, click the three dots on any conversation that mentions a client and look for Lock chat. If it's there, lock that one chat, store your PIN somewhere you'll find it, and remember that resetting a forgotten PIN deletes everything locked. While you're in Settings, open Data controls and check whether Improve the model for everyone is switched on, since locking doesn't change it. If you don't see Lock chat, it hasn't reached your account yet, so check again in a few days. This Friday I'll show you a connector that lets Claude read your iMessages, and I'll start with the privacy rules.

This is my most popular course and the most important thing I run all year. It's four, live online sessions, December 6-12, built around projects you finish as we go.

Declan, from the April 2026 cohort, put it this way: "Highly productive class, project based with real time results. Matt has curated a well chosen path for agents to implement the latest and greatest tools."

Enrollment is open now, and 10% off Earlybird pricing ends this Saturday, October 31.

-Matt