Shopify says traffic coming to its merchants from AI has increased 3x year on year, and orders originating from AI searches have grown at the same rate. Even more interesting, Shopify says new-buyer order rates from AI traffic are twice those of other channels.

For businesses, I think this is worth paying attention to now rather than later. We’ve spent years thinking about how our websites show up on Google. Increasingly, we may also need to think about whether an AI assistant can actually understand what we sell, who it’s for and when to recommend it.

A simple test: ask ChatGPT or Gemini the sort of question one of your customers might ask before buying.

See which businesses come up, and more importantly, why.

If your competitors are easier for AI to understand than you are, that’s probably worth fixing.

Google is giving Ask Maps more agent-like capabilities. Instead of simply asking for nearby restaurants, users can describe what they want and Maps can help find suitable options, compare hotels, surface local events and even help move towards completing things like food orders.

This is one of those updates that sounds small until you think about how we actually use Maps.

Usually it’s five separate searches. Find the restaurant. Check the menu. Look at reviews. Work out whether it’s near where you’re going. Then finally book or order.

The direction Google is heading is basically: just tell Maps what you’re trying to do.

Next time you’re organising a client lunch or travelling somewhere unfamiliar, try giving it the whole request rather than searching one thing at a time.

And if you run a local business, this makes keeping your Google Business Profile accurate even more important. AI can only recommend you based on what it can actually understand about your business.

Adobe has launched a unified plugin that brings more than 70 creative and productivity tools into ChatGPT, including tools from Photoshop, Firefly, Premiere, Acrobat, Lightroom, Illustrator and InDesign.

I find this one interesting because it’s less about another shiny AI tool and more about removing some of the jumping around between tools we already use.

You might start in ChatGPT brainstorming a campaign, then use Adobe tools within the same conversation to start creating or editing the actual assets.

So if you’re already using ChatGPT for marketing ideas, this could be worth experimenting with on the next real piece of work rather than just testing it for fun. Give it an existing campaign image and ask for different social versions, or take a longer video and see how far you can get towards a short-form cut before opening another app.

The goal isn’t necessarily to replace Photoshop or Premiere. It’s to remove a few of the annoying steps between “I have an idea” and “I have something I can actually use.”

OpenAI is rolling out GPT-5.6 Luna as the default model for Free and Go users, alongside unlimited text chats and a new Think option for questions that need a bit more reasoning.

For anyone who has avoided paying for an AI subscription, the gap between the free and paid experience is getting pretty interesting.

The Think button is probably not something you need when you’re asking ChatGPT to tidy up an email. But when you’re comparing pricing options, planning a project or trying to make a decision with a few competing factors, that’s where I’d use it.

One prompt I like for this sort of thing is simply:

Sometimes the most useful thing AI can do isn’t give you another idea. It’s challenge the one you already have.

One of the easiest ways to get more value from AI is to stop treating it like somewhere you have to go and start using it inside the tools where the work already lives.

Gemini in Google Workspace can help surface information from Gmail and Docs, prepare meeting context, summarise threads and turn notes into follow-ups and action items.

This is probably one of the least flashy AI use cases in this newsletter, but honestly, it might be the one that saves you the most time.

Before your next meeting, instead of manually digging through three email threads and trying to remember what everyone agreed to last week, ask Gemini to pull together the recent context and give you a short briefing.

Then after the meeting, use the notes to draft the follow-up and next steps.

Not revolutionary. Just 15 minutes of annoying admin quietly disappearing from your day.

And I’ll take that. 😅

Last week inside AI Systems Lab, we looked at graph engineering, a simple way to stop asking one giant AI chat to research, write, check and approve its own work.

Instead, you break a recurring workflow into smaller jobs, decide what information needs to move between them, and add proper checks and human decisions where they matter.

Think customer support, inbox triage, overdue invoices or major business decisions. The goal isn’t to build some enormous AI system. It’s to take one workflow you already run every week and design a better way for AI to help with it.

If you’re interested in moving beyond prompting and starting to build more reliable AI systems around your actual work, this is exactly what we’re exploring inside the Lab.