in the middle of the month, and i’ve already burned through my a.i. quota at work
Just like in the old days, today I had to manually fix unit tests at work as a software engineer. Recently, the company adopted a monthly AI quota policy—and I burned through all of it. Completely.
The truth is, I think I was more productive without AI (except when it came to tests). That doesn’t mean AI makes us unproductive. It’s just that I got caught up in the vibe coding wave. You know how it is... if I had a choice, I’d rather be living off inheritance than working a corporate job.
For an AI agent to be truly effective, it needs context. It’s almost like working with a junior developer who needs a clearly defined architecture and detailed instructions in order to do good work. I didn’t do that.
First, you provide the full project context (/init in GitHub Copilot). Then you specify the files and code sections that need to be modified. Only after that do you define the requirements—which are not yet the command itself, but rather a discussion, planning session, or refinement phase—if you want to work that way.
Once everything is well defined through that “conversation” with the agent, you let it do the work.
Otherwise, it turns into an endless game of trial and error...
I’m out of tokens.