Like many others, I've been really struggling with all this AI agent stuff. At first I totally rejected their use. Then all the "SOFTWARE ENGINEERING IS DEAD IN 6 MONTHS" posts on social media pushed me towards depression. Then a few truly respectable software developers started being real about it and I finally let go and gave them a try.
At first it sucked because I didn't know what to do. I'd ask tiny questions and expect big answers (thanks to all those stupid ass social media posts like "I JUST ONE-SHOTTED MY OWN SAAS APP, I'M A DIPSHIT!" (I added that last part...).
Weeks went by where I tried more here and there, but nothing worked well. I liked the AI chats (Grok and Claude) for google searching and for iterating on general ideas that were in my head. I even used Claude to help me fix my resume by giving it instructions to take my detailed description of a project I worked on and turn it into a resume bullet point. That worked great (although I'm still looking for a job..).
Then I learned more about "skills" and "commands" in OpenCode and began down a rabbit hole to create a set of them that would help me stay on track with my project goals. I already had a project related to this (https://github.com/Zooce/goal) but I stepped away to see if this could be done with AI instead. This was going well for a few days. I had it create a few skills for dealing with what I called a SPEC.md file that had the project vision and a road map. The skills would instruct the agent on updating the road map as it completed parts of it. It was kind of cool.
But then I realized I was just building a system that would force me to stop doing the work and let the agent do everything. I hated it.
However, it wasn't a waste because when I went back to the drawing board (literally my whiteboard) and listed out the actual problems I had (going from idea to getting started, charting a reasonable path from the start, following a simple process to make progress, and coming back to the project after time away), things started to feel right again. I quickly realized that my `goal` project could address the road map problem with some modifications. A new `/spec` command (and related skill) could help me take my ideas to getting started. New `/remember` and `/handoff` commands (and related skills) could help me come back into a project after some time away. And putting them together would help me make progress even when I have large spans of time in between.
Here's what I realized today. I don't have to let an AI agent do everything. All of those loud ass people on social media can shove it. I don't have to do things that way and AI can still help me out a lot. I want to do the work that I want to do. I like building software. I like writing code - there is something more to the act of writing it. I don't have to write all of it. There's plenty of code that many of us don't like writing - I'll use AI for those parts, for the mundane parts, for the super repetitive parts. I think I'm beginning to find a way to use AI that works for me.
Let's see how this goes.
At first it sucked because I didn't know what to do. I'd ask tiny questions and expect big answers (thanks to all those stupid ass social media posts like "I JUST ONE-SHOTTED MY OWN SAAS APP, I'M A DIPSHIT!" (I added that last part...).
Weeks went by where I tried more here and there, but nothing worked well. I liked the AI chats (Grok and Claude) for google searching and for iterating on general ideas that were in my head. I even used Claude to help me fix my resume by giving it instructions to take my detailed description of a project I worked on and turn it into a resume bullet point. That worked great (although I'm still looking for a job..).
Then I learned more about "skills" and "commands" in OpenCode and began down a rabbit hole to create a set of them that would help me stay on track with my project goals. I already had a project related to this (https://github.com/Zooce/goal) but I stepped away to see if this could be done with AI instead. This was going well for a few days. I had it create a few skills for dealing with what I called a SPEC.md file that had the project vision and a road map. The skills would instruct the agent on updating the road map as it completed parts of it. It was kind of cool.
But then I realized I was just building a system that would force me to stop doing the work and let the agent do everything. I hated it.
However, it wasn't a waste because when I went back to the drawing board (literally my whiteboard) and listed out the actual problems I had (going from idea to getting started, charting a reasonable path from the start, following a simple process to make progress, and coming back to the project after time away), things started to feel right again. I quickly realized that my `goal` project could address the road map problem with some modifications. A new `/spec` command (and related skill) could help me take my ideas to getting started. New `/remember` and `/handoff` commands (and related skills) could help me come back into a project after some time away. And putting them together would help me make progress even when I have large spans of time in between.
Here's what I realized today. I don't have to let an AI agent do everything. All of those loud ass people on social media can shove it. I don't have to do things that way and AI can still help me out a lot. I want to do the work that I want to do. I like building software. I like writing code - there is something more to the act of writing it. I don't have to write all of it. There's plenty of code that many of us don't like writing - I'll use AI for those parts, for the mundane parts, for the super repetitive parts. I think I'm beginning to find a way to use AI that works for me.
Let's see how this goes.