Re: When “technically true” becomes “actually misleading”
This was an interesting new way to look at the people who are perpetually skeptical about AI. For many months now, I was a skeptic. Now, I know that my skepticism was mainly due to the tools that I had access to, and they were simply not good enough. My tools and my use-case did not give me an experience worth talking about. Curiosity persists though. Eventually, I realized that OpenCode is great! It has measurably increased my productivity.
I was able to open three pull requests on Friday. I was finally able to solve problems which had remained unsolved for multiple months because most team members were doing something else and this never came to the top of the “pick up” pile. Before you say anything: The pull requests that I make using OpenCode were not built on full auto-pilot by any means! This is my rough workflow now:
- Every pull request required about 1 hour of “pre-work” where I went through the code (without any AI assistance) and understood the issues and any code that already existed which I could re-use.
- Following this pre-work, the actual AI use was about 10-15 minutes, where I broke down the solution, sent a handful of prompts, inspected the work, and made granular commits which would be easy to review.
- Once the code was done, I spent about 30 minutes more on testing and self-review of the PR that was created, working on improvements (like “Does this class really need to be static or can it have instance variables?”)
So, the cycle takes about 2 hours per pull request, and that’s the formula behind my 3 PRs a day on that day; a particularly productive day.