Ask Paul: November the 13th ⭐

Happy Friday the 13th! Let's kick off the weekend a bit earlier with another great round of reader questions after another momentous week.
"Outage Disrupts Outlook and Other Microsoft 365 Apps"

This is why friends don't let friends use Outlook
🗨️ What prompted you to ask that?
spacecamel asks:

I am loving Antigravity based on your recommendation. I am currently using it to create my annual Xmas card as a webpage that I share with friends and family. However, I have gone through a couple of days of trying to prompt it to get it just right. What is your process for breaking down a problem to write a good prompt? The organization seems different for an AI than what you write for a person. Also, does your wife do it differently?

I haven't really used Antigravity for non-coding tasks yet, but I'm hoping to do so for a few of the items on my AI to-do list this afternoon. But given how good it is for coding, I suspect it will be rewarding. This may seem like a small thing, but Gemini is an example of an app in Windows that can't write to the file system, and one of the tasks I'll be writing about in next week's AI usage update is using AI to help me create a short companion booklet for Eternal Spring with a nice layout, and I had to use a Perl script to generate a file, and that's silly. I feel like Antigravity will handle that directly. I will find out.

I can't claim to be proficient at prompting AI, but it's interesting to me that one of my earlier observations--that it pays to be as specific as possible--may ultimately not be true. But for now, I guess there are two basic schools of thought here. I often provide the simplest and shortest imaginable prompt because AI has gotten better at inferring what it is that I want, and you can then iterate through natural conversation until it's precisely what you want. Or, you can be as specific as possible right up front, which I've also tried. And I'm not sure which way is "better," if one or the other is better in certain situations, or if all this is temporary anyway.

My wife's answer basically boils down to "it depends" because some interactions are one-off requests and some are tied to repetitive tasks you may want to optimize. She uses Gemini as well, and has been building skills in the AI for the repetitive tasks. So she will start by creating a new skill (which in Gemini is buried in settings for some reason), try to use that with whatever project she's working on, and then iterate on it over time. When she creates a finished document of whatever kind, she will show it to Gemini, explain that it's complete, and ask it to update a skill she used to make it. She's a writer, too, and one of the built-in skills is for learning your "voice" when writing so that when AI reworks something she wrote, it will read like something she would write.

She also reminded me that one thing AI is good at is solving problems, which means you can ask it how you might accomplish whatever task--like...