Before You Give Your AI Assistant Too Much
Oct 30, 2026This week the big AI companies all seemed to want the same thing: to know your whole world. Your email. Your calendar. Your business. Your brand colors.
That can make AI way more useful. It also means you need to be a lot more careful about what you hand over.
Let's break it down.
Claude now has an always-on assistant (sort of)
On October 21, Anthropic introduced Claude Duet. It can keep an eye on your connected email, calendar and other apps, put together a morning briefing, work on tasks in the background, and ping you when something needs attention. It can also build reports, presentations and dashboards. Anthropic says it asks permission before sending messages, making payments, signing documents or deleting anything. Right now it's an early preview, by waitlist, for U.S. Pro and Max subscribers. (Source: Anthropic)
It's Anthropic's answer to ChatGPT Dots & Meta's Muse, which I've been playing around with lately.
What this means for you: AI is starting to decide what deserves your attention. So your new job is teaching it what "important" actually means to you. It's pretty damn cool, and the thing that's really stood out to me so far is it seems like it will be a great "chief of staff" type tool. It will need inputs to assess what is important, but so far, it has done a great job.
Gemini wants to learn your whole business
On October 20, Google launched a Gemini hub for small businesses. You can connect tools like Google Workspace, Shopify, Canva and Xero, and upload your website, fonts, colors or brand kit so Gemini remembers your style. It can also turn spreadsheets and receipts into charts and financial models. (Source: Google blog)
What this means for you: no more re-explaining your business every single time you open a chat. That's huge for solo folks.
Just remember a pretty chart isn't the same as good financial advice. Double-check the numbers it started with.
Canva's AI images are actually editable now
On October 22, Canva released Craft, its own image model. The cool part isn't the pictures. It's that you can change individual pieces (move the text, swap an image, fix a color) without starting over. (Source: Canva)
What this means for you: let AI make the starting point. You make the finished product.
ChatGPT added locked chats
On October 20, ChatGPT started rolling out locked chats across Free, Go, Plus and Pro. You can hide certain conversations behind a PIN or your device login. (Source: OpenAI Help Center)
What this means for you: great if someone else uses your computer. But hidden isn't the same as deleted. A locked chat is still stored and still follows your normal data settings. So keep the super sensitive stuff out of it either way.
Before you connect AI to anything, ask:
- What does it actually need to know?
- What can it see?
- What is it allowed to do?
- What has to come back to me first?
One prompt to try this week
Before you let any AI help with a bigger task, paste this in first:
Help me use AI for [task]. Before we begin, tell me what information and access you would need and why. Separate what can remain read-only, what you can draft for me and what must require my approval. Do not send, purchase, publish, sign or delete anything without confirmation.
Holy moly, it's eye-opening to see what it thinks it needs.
The big takeaway
The more AI knows about you, the more useful it gets.
Clear limits are what keep it from getting messy. And organization is becoming increasingly important!
Chat soon!
– Steve
Want to get comfortable using AI every day, minus the overwhelm? That's exactly what AI Unlocked is for: 10 short, hands-on modules in plain English.
This post originally went out to my email list on October 30, 2026.