Jolestar

A bug, and after several modifications with AI, the AI finally concluded that this solution couldn't be fixed and a new approach was needed. I thought of another way to try, and it succeeded. Then in the command line 🎉, at that moment, I suddenly felt a bit of empathy, as if I truly sensed its emotions.

I have been defaulting to using GitHub Copilot for reviewing PRs, after all, GitHub runs it automatically and it doesn’t seem to cost anything.
But in the last few reviews, I started to doubt what it actually “understands.”
For example, a very basic issue: it still considers 1.82.0 to be higher than 1.91.1, which is a common version number comparison error from early large models.
If this were a model problem, it would also think that rust 1.91.1 has not been released yet, which reveals that the agent’s retrieval and real-time status judgment capabilities are also lacking.
Another bigger issue

The annual report generated by Cursor. The update speed of Cursor is really impressive, with several versions released in a single day. Many features make me wish they could be modified, and then the new version comes out and fixes them. However, I have mainly switched to the asynchronous background mode recently. By this time next year, looking at the entire ecosystem, it is likely to be a whole different scene.

After using AI, I can't help but want it to keep working. When I get up at midnight to go to the bathroom and find that AI has finished the task, I just want to assign it another job. But if I'm not careful, I end up toiling away the whole night again.
Later I thought of a way to have the main Claude start another Claude to do the work while supervising it myself. After a lot of effort, I designed a workflow scheme with it, only to find that Claude couldn't learn to start itself; it seemed unable to handle interactive input.
In the end, it said: "I think it would be more appropriat

Encountering a difficult problem, I let AI assist with troubleshooting. The conclusions given by Claude Opts 4.5 and GPT 5.1 Codex Max are different. I copied their replies to each other for debate. Later, I found it a bit troublesome, so I created a document for them to debate by modifying the document. Which AI programming tool has an expert consultation mode that can bring several models into a chat session for debate and finally provide a consultation suggestion?😅

Using Claude-4 for two days, the AI programming ability has indeed improved significantly. Many problems can automatically retrieve code to find solutions, and the need for manual intervention has decreased a lot. I can assign tasks for it to do while eating or sleeping, but it’s a bit costly; if this continues, this month's Transaction History is going to explode 😅.

Let AI write code, I will first have it write a plan to avoid getting lost in the context later on. When AI writes a plan, it tends to include a timeline, as if it were discussing the project schedule with a manager, allowing enough buffer time. For example, if I ask it to improve the testing of a project, it can plan for 5 weeks, but in reality, it can almost get it done in the time of a meal.😅