My on-again-off-again relationship with AI assistants

For the past few years, I have been paying on and off for ChatGPT and Claude. I will read an interesting story about how someone transformed their work with AI, then sign up for a premium plan for one month. I spend a few days actively engaging with the AI and then quickly fall off. I go weeks without chatting with the assistant and then cancel before the month is up.

There is a lot of smoke in the work-productivity AI space. I believe there is (probably) fire there somewhere. But I haven't been able to find it.

Beyond one-off questions that require more nuance than a search on StackOverflow or quick chats about naming things^[1], I struggle to find many use cases for an AI assistant. I don't know if that is a result of how I think and work or a limitation of the product. But it's not working for me.

GitHub Copilot is a little different. It continues to churn in the background while I code, and I frequently utilize its autocompletions. Sometimes the refactors are usable, but others they are way off. I have been experimenting with the more "agentic" features, but I'm finding that reviewing the generated code takes a comparable amount of time to writing it myself. And writing code is more fun than reviewing.

I find AI assistants useful, just less so than other folks online. I'm glad to have them as an option^[2] but am still on the lookout for a reason to pay $20/month for a premium plan. If that all resonants and you have some suggestions, please reach out. I can be convinced!