Happy Australia Day (contention)! I had the joy of taking time off. I also developed on brushstreaks.com. The recent hair pulling issue is data caching. This project was initially crafted with claude code. (Powerful tool for code generation) but I had very little understanding of it's internal workings. This flow is wild to see, especially social media posts emphasising and capable it is at solving problems and sometimes creating x10 the problem.
Back to my use scenario, I look at it from two workflows.
The first, I tried the claude code perspective and asked it to design a plan on tacking the problem and helping me debug. I do admit, when it assisted me, where I asked it to help debug the workflow, I learnt a lot. I could finally understand how the API in a nextjs application can be tested and troubleshooted. I was able to run the command and see the API response. After walking through step by step, we finally saw the issue with the human in the loop and executing the terminal commands. It was a data caching problem.
The second workflow was done by googling "framework" vX.X.X "database" caching and then I had my answer on how to resolve it. Tested it locally, deployed to env and was able to see the difference. It was a one-line change.
The difference, one workflow, I trusted the LLM to guide me. Let it think, let it do the thinking every step. The second, I had to understand the workflow, mentally model it. I was able to diagnose the issue and search precisely the issue online.
Now technically, I used both workflows and arrived at the answer after knowing the problem. And half of the solution is knowing the problem well. I haven't optimised Claude Code yet with skills or post/pre commands. Still learning the basics. But it's clear, expertise and experience still prevail in my workflow.