I built an AI research agent I was sure I’d use every day. I only use it now and then.
It isn’t the only one. The common advice with AI is to turn anything you do more than once into a skill or a workflow, and do it fast. I followed it, and this week I found I’d stopped using most of what I built.
If you’ve set up AI workflows of your own, you’ve likely got a few like this. Looking at mine, the ones that lasted pass three tests. They are:
1. Does it fit how you actually work?
Every workflow starts with a guess about how you work: which step comes first, what slows you down, what you’d happily hand over. Most of us guess wrong, and we only find out after some time has passed.
I guessed I needed research before I write. I was wrong about my own creative process. My ideas come from something I read or something someone said, and I only go looking for research once I have one. I also built a monthly content calendar that added ideas from my analytics. I like coming up with ideas myself, so it was not something I was realistically going to hand over.
What to do: if a workflow sits unused, look at the assumptions behind it before you blame the tool. Move it to the point in your process where you need it, or retire it if it’s a part you’d rather do yourself. My research agent now starts once I’ve written down an idea.
2. Does it start on its own?
If you have to remember to run a workflow, you’ll stop using it. When things get busy, it’s quicker to do the job yourself than to stop and call it up.
And soon it will be forgotten.
I once built five writing skills, one for each stage of a draft. There were too many to remember, so I used each one once and never called them again.
The workflows I use today run without me asking. Every piece of feedback I give AI is saved verbatim and filed as editorial feedback. When I repeat myself, AI flags it, and I decide whether it becomes a rule.
What to do: tie each workflow you keep to something you already do, like sending a message or finishing a draft, so it runs without you calling it. I ask for help in plain words, and it picks the right step.
3. Would it be faster to rebuild it from scratch?
If you keep changing and fixing a setup, there comes a point where rebuilding it from scratch is quicker than fixing it again.
That happened with my first content system. I tore it down in October and rebuilt it in two hours, because by then I knew exactly what I wanted.
What to do: if you’ve lost count of the fixes, rebuild it. What you learnt from the workflows you stopped using tells you what to build.
Here’s how you run the three tests
Copy this into the AI tool you use most:
Look at every AI workflow, skill, agent and automation I’ve set up. For each one, find out: when I last used it, whether it fits how I really work or how I thought I worked, whether it starts on its own or waits for me to remember it, and how often I’ve had to fix it. Then tell me what to keep, what to adapt, what to rebuild and what to drop. For each one I keep, suggest something I already do that could start it.
Run it and let me know if anything surprised you. My DMs are open.