How I used AI this year, 2024 edition

I use artificial intelligence a lot — for work, for home projects, for entertaining my 4-year-old.

But how I use it has changed in 2024.

So! I'm starting an annual tradition. Every December, I'll write up the ways I used AI to make my life easier over the past year, with a bias towards new stuff.

This won't include one-off uses (like generating poems for an art installation). I'll focus instead on broader applications I consistently find helpful.

We'll keep doing this until I get tired of it or the Singularity arrives 😅

Here's the full list:

1. A microtools generator
2. Architecting new apps and projects
3. An answering machine (not that kind)
4. A do-this-thing-for-me helper

1. A microtools generator

tl;dr: AI can make a mini website to solve pretty much any problem you can think of.

Claude has a feature I find absolutely indispensable: It can generate the code for a front-end website, and then render that website immediately.

Maybe that doesn't sound cool on its face. But in practice, it means I can ask Claude to make a bespoke tool for me — and start using it right away.

Here's an example. I recently wanted to generate a bunch of complimentary colors for an app I'm developing. I don't know much about color theory, so previously I would have sought a tool online, or asked ChatGPT to make a suggestion directly.

But this time, I asked Claude to make an interactive tool.

The resulting tool let me play around and pick what I needed, no coding required at all.

(You can use it here!)

This is a game-changer. Not only can I see the logic behind the tool, but I can use it repeatedly without making new AI requests.

And I don't need to paste the generated code into Codepen or something. It's instantly available.

I've used this to help me decide whether I should donate my old car…

… and even how much Chik-Fil-A to purchase for a neighborhood meetup.

2. Architecting new apps (and real-life projects)

tl;dr: Any time I have a big project, I'm using AI to break it into managable chunks.

Back in the day, when I wanted to start a new web project, I'd jump into coding right away. (And waste a bunch of time).

Eventually, I got smarter and started building architecture plans first: What am I trying to accomplish? What data do I need? Which libraries should I use?

Now, I rely on AI for this step. Architecting has become a conversation.

I brain-dump my goals into Claude (it handles projects and code better than ChatGPT), give it constraints, and ask it to pose questions.

What does this give me? Claude builds my data models, suggests new tools, and helps me think through app logic and edge cases.

And this isn't just for coding projects!

I've used Claude to help think through the steps to replacing my house's HVAC system, for instance.

3. An answering machine (not that kind)

tl;dr: I now ask ChatGPT most of my questions, instead of Google — and then ask followups.

I'm not breaking any ground here, but I've reduced my Google / Wikipedia consumption nearly to zero in favor of ChatGPT, especially since SearchGPT became widely available.

In the last 24 hours, I've…

• taken a picture of a flag and had ChatGPT identify it
• generated a memo on the tax implications of donating a car
• got the right Linux command to use to shut down a running process on my Raspberry Pi
• taken a picture of our liquor stock and got a cocktail recipe

I use AI so much, I'm honestly surprised how few of my friends have followed me into LLMworld. I'm constantly quoting ChatGPT responses to them (to the point of being annoying, I'm afraid!)

Yes, ditching Google for ChatGPT is controversial because of hallucinations. The Verge just published "Stop using generative AI as a search engine", highlighting how LLMs make up info.

But personally, I've found ChatGPT much more reliable with Search activated (you can check the links it uses). And when you point out an error, it often "snaps back" and offers a better response. Plus, you can ask follow-up questions, which Google can't handle.

4. A do-this-thing-for-me data helper

tl;dr: No more complex spreadsheets or scripts to transform data — just ask GPT.

I recently sent out Levain Bakery cookies to clients. (Tis the season!) Levain lets you bulk-upload a spreadsheet of recipients — but their required columns didn't match my customer records.

Last year, I would've spent ten minutes reformatting everything manually. This year, I pasted my spreadsheet into ChatGPT and asked it to fix it. Done in seconds.

There's a million little moments like this every week for me. Maybe I need data transformed from one schema to another. Or I need to reformat a URL to remove escaped characters.

In the past, I would have done it manually or Googled for a tool that could do it for me. More often than not, ChatGPT can now just handle it.

Honorable mentions

These are other things I use AI for frequently, but either are too common or too esoteric to include above:

• Create coloring book pages for my kid
• Rating and editing my LinkedIn posts
• Making funny videos on Runway (and now Sora)
• Generating the minutes for my neighborhood community association
• Analyzing transcripts and documents for key points

Things I don't use AI for

Some stuff just isn't for me.

• Generating writing. I like writing and I'm (currently) better at it than the frontier models.
• Autocompleting code: While I like tools like Replit and Cursor Composer to scaffold apps, I absolutely hate the AI-autocomplete functions Cursor and VSCode offer. Nine times out of 10, it's distracting and messes with my flow.
• Suggesting travel plans: AI itineraries feel like warmed-up Tripadvisor listings. Guides like the NYT's "36 Hours In…" series are much better.

Services I pay for

These are the subscriptions I'm shelling out for as of now.

• ChatGPT Plus
• Claude Professional
• Otter.ai
• Cursor
• Midjourney
• Github Copilot
• Runway

I've dropped:

• Perplexity (I don't find it more useful than ChatGPT anymore)

Until next year!

That's it! Hope this was helpful to you. Drop me a note at andrew@mcgill.cool if it was.