The feeling and pitfalls encountered when using Twitter analysis plugin from scratch What did I do A Twitter Chrome plugin that can upload various prompts to form a "character", and then analyze the tweets seen from the perspective of this character. Some feelings and pitfalls: 1. AI can indeed enable people who understand some code and can debug to create a usable tool, but that's all I feel. Mainly for oneself or small teams, it's okay. If you really need to consider concurrency and security issues, unprofessional people may not be able to handle them 2. Running the basic functions first is correct. But when optimizing, it is important to note that it is recommended to feed old code as a "reference material" to AI for it to re plan and rewrite. Don't try to make it slowly change on the original pile of code, the illusion of a big model will really make you doubt life. 3. I have explored and found a more reliable approach: Let AI come up with an overall plan first, and then divide it into modules step by step. Specifically within each module: Let AI come up with a design solution first. -Then repeatedly interrogate it: Did this plan leak anything? How else can it be optimized? -The plan is almost done, let it come up with a detailed execution plan. -We need to continue questioning it: is this plan comprehensive? Is there a pit? -Finally, let the AI write the code and quickly test its effectiveness on its own. Code review is also similar: -Let AI plan a review plan first. -Repeatedly questioning whether this plan covers all aspects and is reliable. -Let AI execute the review according to the plan. -Based on the identified problems, let AI come up with solutions and execution plans, and then promote optimization. A few things that must be noted: - Git！ Once the key steps are completed, make sure to submit them immediately! Otherwise, if the code crashes or you want to backtrack, you will regret it. -Before each step is executed, try to have the AI write the plan and proposal into a document. First, carefully check the document yourself to see if it is reasonable and if it meets your requirements, and then let it work. This is much stronger than simply looking at the code or waiting for the results to come out before reworking. last: To analyze the effectiveness of Meme narrative: https://((x.com))/Pro_0xBi/status/1910900870998630800 Overall testing effectiveness https://((x.com))/Pro_0xBi/status/1910929200371868015