AI handles more and more

This is the fourth article about me using AI from about the end of April, 2026 to about the end of August, 2026 after the writing of Thinkings after being able to code more with AI support.

Choose your AI buddy

If the logics of AI itself change, sometimes really 'bad' things might happen, like very verbose variable naming, e.g. from dataToCreatePostComment to dataToCreatePostCommentFromRequestBody or from commenter to commenterDisplayNameTrimmed, and if you intensely rely on this AI, many new refactors might show up that you suddenly need to take care of, which is like downgrading from ChatGPT 5.6 to 3.5. At the same time, your life can suddenly become very bright by switching from Composer 2.5 to Cursor Grok 4.6 because you find the later writes code following your styles and rules much better. This is definitely not a question about which engine is better and not only a 100% warn of the AI changing its way of doing things but how tight you are combined with your AI buddy.

When cannot solve a bug with one AI for some time, trying to use a different AI might be a good idea. On the frontend, the production server base url environment variable doesn't have www in it, so when connecting to the server, the auth header is stripped because of redirection, which is what I experienced and fixed by working with two AI engines.

AI auditing can be in very good details, e.g. a generateFamilyInviteCode function only needs retrying feature when having more than one million families using the app, which is extremely rare, and a handleCreateFamily function's potential double submitting issue shares the same story. This kind of auditing might be really good at something like finding more bugs before submitting the code because AI is good at building complete logics and finding very edge cases. But at the same time codebase might be becoming bigger and bigger with things that you might not need at least for now? But but are these things actually good and actually people cannot know due to lack of knowledge?

To really consider the AI produced results

For now, AI can give 'too much' code and you may need to refine and refine to get the precise code you need, which might take a lot of energy. For example, your overall security architecture and design is to do daily basic protections and you use a powerful AI that gives you something which is trying to mimic the solution of a banking app with more than ten million users. Will good rules prevent this situation from happening super easily? Is it always true that you do not need anything from a higher level solution even though it's 100% not related to your work on hand or just beyond your project scope or knowledge?

AI has its own way of writing code. Actually if a person does not need to read code at all, no matter how messy (even something like 1 and 0 style of code) the codebase is and the app works (online posts say that there are people who might not read or even write code but use very complete tests and related methods to make sure things work), is it ok? But if this person needs to read code, AI's way of writing code might need a lot of human reviews and refactors, depending on things like company or team or personal styles, specific business or project requirements, time and effort, etc. What's the ultimate future? But even if AI can handle everything, humans still want their own ideas of how to code in their projects?

For my understanding, at least no matter how good AI will be, you can always read the code it writes, and learn 'new' things if you want, and there's always places to add personal 'flavors'. The question is just: do you want to?

Still some questions about thinking and controlling

Even when AI is super powerful, is always using AI reducing our own capability of thinking and affecting other things, e.g. being more and more reluctant to read news, articles or books but always letting AI give us summarized knowledge?

AI is good and giving more and more solid suggestions but at the same time I need to learn even more than before and the feeling is becoming stronger and stronger that I don't know a lot of things and it's hard for me to control my codebase gradually. Then I need AI even even more to fix potential bugs if anything it suggests goes wrong so whether in the future AI will just handle everything And reviewing code is finally more like writing letters versus sending emails?

The tech world develops so fast but there are many places that only 'simple' code works can have big impacts, e.g. poor countries or regions in the world or even not that developed places in the States where many daily things or works are not automated or covered by code, non-profit jobs that are actually very beneficial to the great nation or certain peoples, you name it.

And finally, 100% home education might be very great again?