I’ve been trying to learn a surprisingly difficult new AI skill: not doing things myself.
Outside of work, I’ve been automating a pet project that requires me to create and post content to Instagram every day. For a while, that meant spending a lot of time in Canva—moving things around, changing colors, resizing text, adding animations, and making dozens of tiny decisions.
There was some value in doing it manually at first. I learned how Canva worked, figured out what I liked, and discovered its features and many little quirks. Then my husband suggested something that sounded obvious but felt strangely uncomfortable: stop touching Canva. Instead, tell Codex what I want and make it do the work.
My first reaction was basically: but it would be faster if I just fixed this one thing myself. And sometimes, in that exact moment, it probably would be. Opening Canva, nudging a text box ten pixels, and moving on takes thirty seconds. Explaining the change to AI might take longer.
But if I keep making every thirty-second fix myself, I’m also guaranteeing that I’ll still be making those fixes tomorrow.
So I started forcing myself not to. I started by writing a 15-page google doc explaining how I want my posts created: layouts, fonts, spacing, animations, naming conventions, what should change each day, and what should stay consistent.
The result isn’t as creative as what I would make if I designed every post myself. But that has started to feel somewhat beside the point. I’m no longer trying to make each Canva post perfect; I’m trying to build a system that can make the next hundred without me. And now even that Markdown file is being maintained and updated by Codex.
Once I started looking for it, I noticed the same pattern elsewhere.
At work, we’ve been updating some of our legal templates. I would leave comments in a document with changes I wanted made. My instinct was to collect those comments, translate them into a clean set of instructions, and then give those instructions to Claude.
Our General Counsel did something much simpler: he told Claude to read my comments and make the changes.
Of course!
I had accidentally created a job for myself that didn’t need to exist. We’re so accustomed to being the person who transfers information from one place to another, reformats it, summarizes it, enters it into another system, or clicks the buttons required to make something happen that we don’t always notice how much of our work consists of being the intermediary.
Increasingly, I find myself asking: why am I in this step at all?
I think that is the bigger shift I’m starting to notice. At first, I used AI mostly to help me do things faster. Draft this. Summarize that. Help me figure out this formula. But more and more, I’m trying to use it to remove entire steps that I used to perform myself.
That requires a different mindset. It means resisting the temptation to jump in every time something needs a small fix. It means spending more time upfront explaining what good looks like. And it means getting better at reviewing the work closely enough to catch what is wrong, refine the instructions, and improve the system over time.
For a long time, being good at software meant knowing how to operate it. I’m starting to think the next version looks different: knowing what outcome you want and resisting the urge to grab the mouse and do the work yourself.
If you’ve made it this far and are curious about my pet project, check out www.ThatPetName.com (and follow @ThatPetName on Instagram!)