I can only talk about my own experience using AI for work but when you have it do the project documentation at the same time it tends to be more clear and focused. If anything, I’d ask it to create comprehensive documentation so that IT can understand it better later (because the AI has no long-term memory).
I have yet to find AI code that’s hard to understand nor so complex that I couldn’t figure it out and having it document things as it goes is really helpful for me! But then I’ve learned to spot when it’s going into a rabbit hole. Which, frankly, I think is the skill to hone (knowing when it’s full of shit)
The danger behind AI is that you are not understanding your own code at the end and then also: do not know what the code is doing. When I use AI for Java I can understand the code at the end. If I would do this in let’s say C# tht could be more critical for me to understand the hole code possibly. It is dangerous to use cause it makes everything more complex. Firt learning the language you are working with and I mean: really learning and being able to work with it intuitive. That’s the fundamental need to work with the AI cause you need to be able to understand every line of code. Otherwise you may end up in a project which was becoming something else in between.
Currently what I am building are MVPs so maintaining them will not be an issue at this stage.
Of course I will have to eventually learn Next & Expo but that is in near future and once a MVP is approved and a full fledged product has to be built. Till then I can just relax and enjoy using AI become more productive.
Actually adding features is a breeze. If you have not used AI for coding then it may feel overwhelming.
After using it I have come to the conclusion that AI seems to work best in following languages:
Java
Python
JavaScript
Typescript
Flutter/Dart
I may be wrong in my assessment but the result that I have managed to get in all these are just mind boggling.
Besides documentation I have come to appreciate its capabilities to suggest complete database schema for a project based on implementation plan.
I have used Supabase in many web app as back end. In fact for my team Supabase is the default back end. But I had always avoided using `Enumerated Types` but in the mobile app project that I am currently building, AI suggested that I use Enumerated Types. Now using Enumerated Types feels so natural. Whenever I want to provide users with finite choices and force the RDBMS to store only a valid value from the list, Enumerated Types comes to my rescue.
I also love the RLS generation feature. AI can study the schema of any database and generate a correct RLS policy. Before we used to labor on writing the policy.
Please don’t feel depressed. The practicality is that whether we like it or not, AI is the future productivity tool and is here to stay. The faster we adopt it the better for us.
Let me share one more experience. We have loads of web app that are built in PHP. One of my new team member got Claude Code to convert a project to MERN stack and the project is working. Of course it has not been tested extensively but all the features that I tested are working!
Yeah, tht’s what I meant. You needed an AI account which was not free. You did not payed for it. That’s not meaning that others would have it for free.
No, you’ve made a wildly misleading claim and I’ve asked Norman to review it. I am looking for alternatives, but the reality is your company spends $200/month on AI that they’re letting you use. That’s 20% the cost of Xojo, nowhere near free.
To be fair, he does say “almost free”. And (supposedly) the full $200/month is not being used solely on this project. But your point is valid.
@YogiYang, what would substantiate your claim is to tally the actual cost of the Claude Code resources you’ve used for this project so we can all ascertain how close to “free” it really is (the threshold for that will be different for each person).
No, almost free means nearly free… 10, 20 or 50 Bucks. Not 2400 Bucks a year to speak truths. That’s what yu have to pay to use it. And that is NOT cheap at all. I have no Idea what for you is nothing. But even with my incomes I would say: 2400 per year is money.
By the way: building a mobile Android and iOS App for free is possible. With flutter/Dart. With CodenameOne. And several other solutions. For no upfront costs. If you have no macOS computer I would recommend you CodenameOne. Cause they provide you a BuildServer you can use until a jar size of 8MB. If you have a mac, you can build entirely local for Android and iOS without using their Buildservers. Have fun. THAT’s for FREE. And COPILOT gives the right HINTS for programming. And if there are questions: ask.
I have categorically mentioned that for the mobile project I am using the free tier of Google Antigravity (Gemini) but for official work we are using Claude Code Max plan. And there are 4 other developers besides me using Claude Code at work.
And the title of the thread mentions …almost free.
I repeat I am not using Claude Code for the mobile app development. I am only using Google Antigravity. I have not spend a single penny till date in building the mobile app.
In my first post I am just sharing my experience that Gemini is very good for building no-nonsense UI/UX while Claude Code is very good for coding. I have learned that Claude Code tries to be over smart at times and build things that are never asked for. This probably may be for increasing token usage count.
I only react to that: I have a problem with a project and I do not know where to look. And this is a rewrite some 10 years later of another project: the goal was to make it more “understandable”, faster and add some new features: I failed.
What about AI ? It revive my bad memory when I forgot how to do something: I ask Gemini, read its explanation and eventually take its snippet or the shared code inspire me to write my own / corred the shared one to fit my needs.
45 years of experience heren (I am old and forget things too fast).
