Gemini with Personal Intelligence is awfully familiar

If that all sounds familiar, it’s because Gemini already offered the option to hook into your Workspace apps. But it required more work on the user’s part — I always found I needed to explicitly ask it to check something in my email or on my calendar if I wanted it to use those as sources. Now, if the prompt seems like it merits a trip to your inbox to look for an email about a concert ticket, it will do so of its own accord. That’s kind of huge. If you have to be specific with every single prompt and babysit the AI, then it’s no more useful than the timer-setting robot assistants we’ve been using for the past decade.

Gemini offers some suggested prompts to try once you enable Personal Intelligence, like having it recommend books you might like based on your interests. The titles it suggested to me were annoyingly spot-on. Another one of these conversation starters resulted in a lengthy chat with strategies for dealing with the lawn in my backyard, which I hate and the crows are picking apart anyway. Gemini offered some native plant options to consider, added reminders to my calendar based on the plan I settled on, and put together a shopping list in Keep that I could bring to the hardware store. Even a couple of months ago, Gemini would routinely fail when I asked it to complete tasks like “Add this to my calendar,” so that’s a significant leap forward.

The thing is, Gemini gets out over its skis in other ways. I had it brainstorm some new bike routes, asking it to incorporate a stop at a coffee shop. It obliged, and its high-level recommendations were good, though it struggled with the finer details. Trying to nail down specific routes was painful; it would give me a link to a route it claimed to have created in Google Maps, but clicking through to the map showed me a different set of directions. I’m also not convinced about its plan to send me through the woods on some unpaved trails, culminating into a left turn cutting across several lanes of traffic on a busy road, so I’ll probably stick to the routes I know.

That’s the problem. Gemini can analyze my interests and make some pretty good guesses about what I’d be interested in; it’s the details where AI gets lost. I asked it to look for some neighborhoods I might be less familiar with to recommend for an afternoon outing to take pictures and (naturally) get a coffee. It used my personal data to correctly work out that I’d previously lived in Ballard and shouldn’t include it as a recommendation. The overall list it came up with is solid; the specific locations it recommended weren’t always right.

It claimed a restaurant in South Park was in Georgetown, said I would find a Caffe Umbria in the Old Rainier Brewery building (none exists there), and heartily endorsed a T-shirt shop that is quite obviously closed based on its Google Maps listing. I had to do enough fact-checking and reprompting that it all started to feel like more work than it was worth.

That might be Gemini’s biggest immediate challenge. A year ago, it needed a lot of babysitting to get to the personal information I needed, and it got stuff wrong regularly. Now, it can do the personal stuff reliably — but getting details wrong is a pretty big bug. You only need to show up once at a vacant storefront to decide you’re done using Gemini. That’s not even touching the privacy aspect of it all. Gemini referenced my husband and child by name in one of our conversations. It’s one thing to know that that information is trivially easy to find with access to my email and calendar; it’s another thing to hear their names out loud.

Misgivings aside, I think the inclusion of Personal Intelligence has increased the scope of what I’ll use Gemini for — but only slightly, and I wasn’t using it a whole lot in my day-to-day to begin with. I have a schedule for my yard work and a list to take to the neighborhood nursery, where I’ll ask an actual human if I’m on the right track. Maybe doing that initial planning with Gemini will be what helps me feel just confident enough to get started, even if I end up course-correcting down the line. That’s not a bad tool to have. But you can bet I’m going to be watching my step on whatever path it recommends for me.