3 Things AI: Issue #9

Welcome to Issue #9 of 3 Things AI - a biweekly* note where I share three practical or interesting ways I used AI.

Thanks for all the great feedback on the last issue, and if you didn't check out the AI-prompted Spotify playlist that I shared with "Focused Work Flow" tracks – check it out here!

Tax season has a way of making me feel like I'm behind on things I didn't know I was behind on. Add in a stack of rejected insurance claims and a new voice tool I can't stop using, and that's pretty much my last two weeks. Here are three things.

My son is in college and TurboTax kept telling me he didn't qualify for the American Opportunity Tax Credit – the federal tax credit for undergraduate education expenses. I went back through every field multiple times. Nothing looked wrong. But it still wasn't giving him the credit and I had no idea what I was missing.

So I set up a Claude Project and fed it the full picture: his school, year in school, dependent status, what we'd paid, our income range, where things stood in the filing (I redacted anything sensitive from uploaded files like SSN, etc.). The reason I used a Project specifically is that Claude Projects hold context between conversations – so I could lay out everything once and keep drilling in without re-explaining the situation each time.

It looked at the whole filing in context and spotted what I'd gotten wrong. There was a specific field I'd answered incorrectly that was quietly disqualifying the credit – not obviously, just incorrectly. Claude explained exactly what the issue was, why it mattered, and what to change.

I went back into TurboTax, fixed it, and he got the credit.

(Reminder: always verify tax stuff with a real professional. But for diagnosing why something isn't working the way it should? This is exactly where having full context in a Project pays off.)

At some point last year I saw a provider a handful of times and went to submit those claims to my insurance for reimbursement. They came back rejected. The superbills – the itemized receipts providers give you for insurance purposes – were missing required fields. Some were missing diagnosis codes. Some were missing CPT codes (the standardized codes that describe what service was provided). In a few cases the code was technically there, but buried in a table instead of appearing directly in the service line, which my insurer apparently doesn't accept.

I had five statements, each with its own specific issue. Writing a clear correction request to the billing team felt like the kind of task that sounds easy until you're actually doing it carefully, because medical billing has its own language and I didn't want to send a vague "can you fix these?" email that just created more back and forth.

So I uploaded everything into Gemini. I explained the situation – insurer requirements, what had been rejected and why – and asked it to review each statement, identify exactly what was missing or incorrect, and organize the findings into a spreadsheet with a row per statement.

It handled this really well. Within a few minutes I had a clean, organized spreadsheet. Then I asked Gemini to write the email to the billing team using the spreadsheet as its source. It drafted a clear, professional request that walked through each issue specifically and told them exactly what needed to change on each statement. I sent it almost word for word.

(I'll be honest: I usually like to do my own writing. But for anything involving medical billing, IT requests, or other contexts where the language needs to be precise and I don't really speak the language – I'm totally fine letting AI write it. This is exactly what it's for.)

The billing team still hasn't finished the updates, to be honest. But when I followed up, they knew exactly what I was asking for. That part, at least, was not a pain.

I've written before about being all in on voice as the future of how we interact with these tools. I tried Wispr Flow in Issue #5 and was impressed. Since then I've switched to a tool called Monologue, from Every.to, and it's become the one I reach for constantly.

The basic idea is the same: press a button, speak, and it turns your words into polished, formatted writing. But Monologue is really smart about context. It figures out whether you're in an email and formats accordingly – full sentences, punctuation, proper paragraphs. Switch over to a text message and it relaxes. More casual, less structured. It picks that up automatically without you having to configure anything.

It also does a great job of just getting out of the way. Music playing in the background? Doesn't matter. You can make corrections mid-sentence and it cleans them up. It cuts the ums and the likes and the false starts and just gives you what you meant to say.

I use it for AI prompts, work chats, emails – it's just on all the time now. And unlike typing, which sometimes slows me down when I'm trying to think through something complicated, speaking lets me just go. Typing creates a slight delay between the thought and the word. Speaking doesn't.

It works on both Mac and iPhone, which is huge. A lot of these tools are desktop-only, and then your phone is still a typing situation. Not this one. Monologue actually follows you.

If you're curious about trying voice for your AI interactions, Monologue is where I'd start right now. It's part of an Every.to subscription, and there's a free trial. Try talking your way through something you'd normally type – a prompt, an email, whatever – and see how it feels.

__________

*It’s official! I moved the cadence to every other week. I suppose if I just had AI write this I could crank out more, but I enjoy the act of writing and want to give myself the space and time to share my thoughts.

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 every other week direct to your email. If it’s useful, you can subscribe here → https://3thingsai.beehiiv.com/subscribe