Let people make TODOs, not codes.
I think many people do this without clearly verbalizing it (me too).
Some people can't conceive that it should be possible.
Some people think, "Of course I can do it," and then they try it, and then they do it.
The power that should be able to do it.
OKtamajun When you get to this phase, have Cursor Composer have o3-mini scrutinize the entire code (Ctrl+Enter) to find a structural level I would then ask him to write a TODO for the coder without writing any code, and then have Claude 3.7 Sonnet code it after he spits it out. You can make it even bigger by switching to a style of
Also, there is a certain pattern that Claude tends to use for ad-hoc solutions, and if we can see a pattern of how to deal with each one, we can move forward. For example, giving the URL of the library's GitHub, making a turn to write down the code design in md, or relying on Claude 3.7 thinking. It becomes more and more like team management, and the main human task is to make sure that the team has the right information, and that they have optimized and synchronized documentation in place frequently.
Of course, there are times when none of the patterns can solve the problem and it is necessary to eta or fix it by hand, but I have the impression that it is easier to move forward on a large scale than before if you have many options with the Cline and Cursor two-fisted approach.
>AmadeusSVX: My impression with Cline is that you can make something pretty simple, but once you hit the wall and LLM itself starts to do ad-hoc solutions, the project seems to start breaking down more and more (it keeps adding useless changes).
At that point, if you can't figure out and solve the problem to some extent on your own, you're going to be stuck.
Well, I think Cline and Cursor are too close to each other in their choices.
I'm Devin, Roo Code and ChatGPT (o1 pro / o3-mini-high)
AI generates TODOs.
---
This page is auto-translated from /nishio/コードではなくTODOを作らせる using DeepL. If you looks something interesting but the auto-translated English is not good enough to understand it, feel free to let me know at @nishio_en. I'm very happy to spread my thought to non-Japanese readers.