Practical AI usages and my understanding of its current future

Two months have past since I wrote Me and My AI 101 during which I had much more practices of using AI than before. These practices are both interesting and very enlightening to me.

Two cases

I have been writing posts/notes on RedNote (a Chinese social media app, very popular these days) about beer styles according to BJCP Style Guidelines since middle of last November. Everyday except weekends I write one post about one style covering things like style introduction, category & parameters, brewery data, declaration, tags and so on. At first, I selected key descriptions of a style from the guide and led AI translate them into Chinese (AI will try to connect the descriptions even I forgot to ask about this in the instruction) and for brewery data, I just led AI list brewery names behind specific beers of a style. Later, AI helped transform the guide into JSON data, it'll find all raw materials and write introductions according to instructions like picking up 1 or 2 key descriptions from categories like aroma, flavor and mouthfeel and writing the summary, and besides just brewery names, AI is helping find the exact beer links on their official sites.

We had a bottle sharing event around the end of last month. I'm always thinking of how to build good beer lists (very briefly, a long image with beer info like label, name, brewery, statistics, description, etc.) automatically and more efficiently. Before, I'd manually get all needed beer info from Untappd and design and build lists using Figma. With AI this time, I gave it all the Untappd beer links and design instructions like the style of each beer item (label on the left with 6px margin to the beer info section with brewery plus name, statistics and description from top to bottom), and AI helped generate HTML first and then convert it to a JPG image with all the other specific tweaks much easier to be implemented.

An idea to realization age

The overall ideas of these two cases are quite simple but I still consider them as the miniature of our future (coding) world. If you have an idea, write down all the logics of how it will run and know where the data is from, AI will help you visualize the final results with great potentials and advantages of building even more functionalities and much more easily tweaking the already existing works in batches, which is not only useful for defeating the most difficult question on LeetCode or HackerRank but also when you are working on a dedicated small, complicated but key feature of a trading system at your company.

If the tech world keeps evolving like this, everybody might have their own applications in the future and the communications between us are much more supported by our applications that represent the crystallization of a individual person's thought and action. If this future coms true, selling APIs (the capabilities of one app communicates with the other) will be a big market for nowadays' big companies.

Still some current questions before the real future

Current pricing model of using AI is more like paying for how much mobile data you were using during the 3G or even 2G age, for example, Cursor has Pro ($20 per month) or Pro+ ($60 per month) plans, which doesn't mean the usages are totally covered and the only differences are some advanced models that lower plan subscribers cannot use but all plans still have usage limits especially if you use 'Opus 4.6 Max' to just change the font of all beer items' descriptions, you'll quickly run out of quota even if you are on Ultra plan.

So where's the boundary of using and not using AI? I hope the answer to this question is definitely not only about saving cost. For example, using more/only auto completions that are free to use (totally covered by the plan) can address much/all cost concerns. However at the same time, if you use more auto completions, you might be actually more close to your code than just letting AI build everything for you. The current development speed of AI will gradually ease the worries that AI will definitely make mistakes and handling big tasks to AI and trusting AI will ultimately let you or your company pay back in the future. The question is actually whether using too much AI will influence your original capabilities of coding or even thinking.

Back to practical examples again

For me, I won't resist using AI anymore but am actually struggling for some things when having these great AI powers on hand. What's a better workflow of working with AI for daily works and for instance, one specific issue is how to write and manage all communication/instruction docs (now plan is very important to use AI well) that are the very key connectors between me and AI? What's the ultimate good or perfect code of fulfilling certain tasks like building database RSLs, dealing with API requests concerning loading, error and data status, building shared component for frontend (used for not only one project), building shared logics and utilities for backend and so on? Because only with nowadays' AI, it already became much easier to optimize the whole database even for a small but useful style change (sleeping in a tech debt tickets for years), which I think is not something about OCD or clean code freak but people's ambitions of building things great when getting much greater tools than before.

Having brilliant ideas and especially some real business that you can contribute you coding talents to might again become the key factor of being success or finding meaning of life in the near future. Some say current vibe coded apps are more like just CRUDs in different colorful clothings with similar core functionalities. But at least I feel the days are gone when I could not fail asleep and kept thinking how to solve that one specific issue which swallowed my whole day times without any progress.