3 Things AI: Issue #4

[2026 - Week 3]

Welcome to Issue #4 of 3 Things AI - a weekly note where I share three practical or interesting ways I used AI this week.

I was fortunate to have a nice break from work over the holidays. For some reason, this year felt particularly difficult getting back in the groove (I also heard this from many others). While it pains me to say it, I did not find a way to solve this using AI. Maybe next year?

In the last issue, I also wrote about my AI-powered plans to try ice fishing. Alas, it warmed up here in northern Vermont, so stay tuned, hopefully I can try again soon.

That said, I did find a few small ways AI was genuinely useful this week.

We got a new made-in-Vermont woodstove for our basement, which is partially finished but 100% unheated (it sits about 50 deg for almost half the year and it takes at least two blankets to watch a movie). The very first fire needed to be a low temperature fire - just newspaper and a few very small kindling pieces - to “break in” the stove and not crack the soapstone. So I tried that late one night, when the family was all asleep, and it was about 15 degrees outside. I could feel cold air rush in when I opened the stove door, but figured the fire would overcome that. Experienced stove owners are laughing at my naiveté now, and I should have asked AI for advice!

Well, I lit the paper and immediately it all caught fire and smoke came straight out of the stove and filled the basement. Now I can assure you, I did not grab my phone and start using ChatGPT. I made sure the fire was out, opened the few small basement windows, and made sure everything was safe. At this point if you had a spouse, a parent, a neighbor, or a friend to call and get advice, that would be great. I did not, so I put my earbuds in, opened the app and started blabbing away.

My experience affirmed two new attributes of this early AI era: (1) searching in Google will soon be antiquated and in fact is very robotic (more on that below), and (2) I believe speaking with AI will become one of the predominant ways to interact with these tools (I already expressed my love for the voice feature in the first issue).

I needed to get the smoke out of my house, minimize soot or odor damage, and do this all without waking up my wife (who had just broke her toe so could not help!). Imagine Googling at this moment something like “basement filled with smoke what to do next?”

As the screenshot below shows, you get a reddit thread and a bunch of articles that are somewhat relevant, but not for that exact moment. This is where AI actually allows us to be ourselves, to talk like a human and get the info we need at the time we need it.

It also doesn’t get annoyed when you ask it the same thing ten times but in different ways, or just get sidetracked fixing one problem and ask it to remind you of something it told you two hours ago. Few neighbors, friends or parents would have that patience with you!

I literally conversed with the app while taking action, for well over an hour, and described my situation, if I should use fans in the windows, whether or not I should blow air in or exhaust smoke out (turns out I needed to blow a lot of fresh air in), and how to start deodorizing the space (wiping surfaces, bowls of vinegar and lots of baking soda).

I got it all cleaned up, though it took a few days to get the smell out. But I quickly learned how to pre-warm the flue and ensure I had a good draft for next time, and I’m happy to report that the stove is working great and everyone’s a lot happier in the basement!

I have used Gmail since 2004, when a colleague invited me at the time that new users could only invite a handful of friends, and there were waitlists to get an account. That’s a lot of history in that inbox (and I’ve never deleted anything). And as good as Gmail and Google are at searching, anyone who has tried to find old emails knows it can be super painful (true in any email service).

What’s funny is that the term “AI” triggers people to think about robots, but really, the way we use the internet today is very robotic. When you search for something, you have to string together key words. With AI, you just type or talk in a natural manner. You literally tell or ask the tool what you want to know. I said last time this can give everyday people more agency to create, and to learn, but I’m more and more convinced that it may also reward those who aren’t technical programmers, but who can express very human thoughts, feelings and ideas.

So I paid $2.99/month for a 3 month promotion (then it becomes $20/month, so we’ll see if it’s worth it) to have the “pro” version of Gemini and connect it to my Google account and all the associated apps - including Gmail, Drive, Docs, Sheets, etc.). There is a free version of Gemini already available on your Google account, but I wanted to play with the full feature set. I’m still early and my first impressions are it’s a bit clunky, but super promising as they iron some things out. Here’s a quick snapshot at some early experiments.

I first searched for all emails from my Mom (who passed away in 2006), and it listed them all and summarized the content. Sure, I could have just searched by her email, so this wasn’t mind blowing, but the summary of all those emails was pretty special. Definitely first AI-induced tearjerker for me. You can also search for things like “summarize all the quotes I got to replace the garage door but don’t include Acme Garage, and list the estimates from lowest to highest.” The results are amazing and you can imagine how much better this is to probe your inbox.

I also used a natural language prompt to create a ”scheduled action,” which is something Gemini will do on a regular basis and provide me with the result. Local readers of mine and others in Vermont use Front Porch Forum to get very local (your town, or neighborhood in some cases) listings for free items, things for sale, complaints about noises, lost pets, wild animal sightings, selectboard updates, and so on. The problem is if you don’t read the email digest you may miss something important. So I gave it this prompt: “Every morning at 9 AM, check my Gmail for any new Front Porch Forum emails and give me a summary of anything that is for sale. Do not include free offerings. If nothing is for sale, then just say that clearly.”

I could have specified “hockey skates” or “kid’s bicycle” or really anything. Here’s the report below. I corrected and massaged the output over a bunch of prompts, and it is somewhat stubborn (e.g., the “source 1” links to the original email but it used to show an icon) and it’s not as fluid as ChatGPT, but ultimately I get this nice summary update every morning at 9!

The possibilities are endless (while writing this newsletter I just created one that summarizes the ski report email from the local mountain - see below), and if you use Google/Gmail for work, this will be very powerful.

At work, I needed LinkedIn profiles for a large group of people (over two dozen) from multiple companies for a bid we were submitting. In the old days, this would have meant opening profile after profile, searching names, double-checking titles, and inevitably missing a few. Instead, I used AI to generate the full list of LinkedIn addresses in minutes. I just put all the names in one column in a spreadsheet, and then copied and pasted that whole column into ChatGPT, and asked for it to return a three column table: Name, LinkedIn profile, and any notes on certainty or confusion finding that profile.

It turned what used to be a tedious, error-prone task into something fast and almost boring. A small use case, but a good reminder of how much friction AI quietly removes from everyday work.

__________

If you have feedback, questions, or an AI tool you’re enjoying, just reply. I’m always curious what others are using.

I send this once a week. If it’s useful, you can subscribe here → https://3thingsai.beehiiv.com/subscribe