You don't need to keep up with every AI tool

Watch the video

I use AI every day. I have since the first version of ChatGPT. I love what I can do with it. Trying to keep up with everything around it is the stressful part.

Some software videos feel like watching MrBeast, except the subject is accounting software. The frantic editing. The shocked face. The suggestion that your job is over if you don't watch this right now.

It's accounting software. We can relax.

Here are three things that have helped me get more out of these tools and enjoy using them.

Mute the pressure#

I'm putting this first. Nobody needs to stay until the end to find out how to make their feed less annoying.

Mute the accounts that repeatedly make you feel behind. You can look up a tool when you have a reason to use it. You don't need a dramatic introduction to every release.

But wait, Parker. You just said it's hard to keep up. Now you're telling me to mute the people explaining it?

No, no. It's fine. Chill.

Some of that information will matter to you. A lot of it won't. Someone making a video urgently does not make their subject urgent for your work.

If a creator does something similar to you, their experience can be useful. Watch the review. Learn something. Then get your own experience with the tool. Their business, starting material and idea of a good result may be different from yours.

Give it a real job#

A pelican on a bicycle can be a fun test. It doesn't tell me whether a model will help with the software I'm building. If your work involves illustration, you might care about a completely different result than I do.

Pick something you actually need done. Before you start, describe what a useful result would look like.

Did it do the thing? Did it do it well? Did it even know what doing it well meant?

That's a small evaluation. You don't need to make it more complicated than the task requires.

I'm building Little Worker, my software for managing tasks and working with AI models. It isn't released yet. Recently, I tried a new model on some product changes and liked what it did.

But product work was only part of the job. My setup also runs checks intended to improve the code before it reaches users. The model I liked for product work spent far longer on that stage than I wanted. The run lasted 47 hours.

Someone could reasonably say, "But it's making the code better."

Sure. That still wasn't the result I wanted. I wanted the work ready to go in a much shorter period. Continuing to improve something indefinitely wasn't my definition of success.

That experience tells me something about the model, the task and the way I set it up. It doesn't tell me the model is bad at everything. I liked it for another part of the work.

Check the whole result#

Look beyond the first impressive output. Did the tool finish the job? What did you have to check or fix? How much time did the whole thing take?

If you're comparing cost, use the usage or billing records you can actually verify. A result that's slightly better might require enough extra time or checking that you don't want to use it again.

I did a version of this to myself with video editors last year. I kept trying different tools while making videos every day. A lot of them did much of the same work. Learning another editor was taking time I could have spent making the video.

Sometimes switching helps. Sometimes the thing you already use is doing the job.

You don't need to try everything. Pick something that matters to your day. Give it a job, check what happened and form your own opinion. Keep what helps. You can change your mind after the next task.

What have you tried that was a total flop? What worked so well that you kept using it? Tell me what you asked it to do. That's the part I'm interested in.