I worry that because we are now able to instantly produce a bunch of JS to do X thing, we will be incentivized not to change the underlying tools (because one, only AI's are using it, and two AI's won't know how to use the new thing)
I worry this will stall progress.
It's not really that different from taking your 2022 car to a shop to adjust your camless engine, and assuming everything's fine, but not having a clue what they did to it or how to fix it if the engine explodes the next day. You can't even prove it had something to do with what the shop did, or if they actually did anything at all. They probably don't even know what they did.
It won't stall progress for clever people who actually want to figure things out and know what they're doing. But it will certainly produce lots more garbage.
That said, the game is impressive for something stitched together by an LLM.
You still need paradigmatic shifts in architecture to enable delivering scale and quality from a smaller amount of materials, and it has not made a dent there, yet.
The standard for new frameworks won't be "does this make humans more productive using new concepts". It will be "can I get an LLM to generate code that uses this framework".
Gemini in particular is really good at this
Obviously it will do even better if you give it the full documentation but it doesn't do that badly in general with the language when you provide a sample app where it can basically just pick up the patterns of the language/framework.
I also find it interesting that the ai code submissions like this one are generally vague about the process.
This seem to be created using Cline on VsCode, prompting to Gemini 2.5 Pro using OpenRouter.
The commit history implies that a crude version was created by the LLM using an initial prompt and then gradually improved with features, fixes etc. Assuming ongoing conversations with the AI agent.
All code in a single index.html file which might not be great for human coders but who cares to be fair.
All in all, a prompt history would be really educational if anyone thinks about doing something similar.
As we start to develop AI first programs, I believe we will need to start connecting LLM conversations to code, not only for educational purpose but for maintenance as well. I'm currently experimenting with what I call Block-UUIDs, and they are designed to make it easy to trace LLM generated code. You can see what I mean in the link below, which contains a simple hello world example.
https://app.gitsense.com/?chat=7d09a63f-d684-4e2c-97b2-71aa1...
Something worth noting is, you can't expect the LLM to properly generate a UUID. If you ask the LLM, it'll says it can, but I don't trust it to do it correctly all the time. Since I can't trust the LLM, I instruct the LLM to use a template string, which I can replace on the server side. I've also found LLMs will not always follow instructions and will generate UUID and how I handle this is, when the LLM stops streaming, I will validate and fix if needed, any invalid UUIDs.
How I see things playing out in the future is, we will alway link to LLM conversations which will give use the Block-UUIDs generated and by looking at the code, we can see what Block-UUID was used and how it came about.
Full Disclosure: This is my tool
Additionally, you get more directly usable text out of a 'git blame'
Ultimately the code needs to stand alone, but if you discover that a specific version of an LLM produced vulnerable code, you have no recourse but to try again and read the generated code more carefully. And reading code carefully is the opposite of vibe-coding.
I wonder if there is some tool support yet that supports that.
i.e. who cares when the LLM starts giving up when the file is too big and confusing?
i.e. who cares when the LLM introduces bugs / flaws it isn't away of?
If the AI fucks up then its a lost cause. And maybe you’ll better off create a new version from scratch instead of trying to maintain one when LLM starts to fail. Just ask Gemini 3.5 this time around.
The AI can write obfuscated code. Name all variables from a to z. Even emit binaries directly. Who cares if it works?
AI won't cope well as that file gets larger, or best hope that experimental diff feature is working well. I find stuff really breaks if you don't refactor it down.
Could not parse specific advice due to formatting issues in the AI response content.
“I had a long flight today, so in the airport I coded something”
we cam start the discussion now i.e.: is using cursor still coding ?
is a senior team lead coding when the only thing he does is code reviews and steering the team?
Happy to give more details if there's any way to get in touch outside this thread.
I had something like $500,000. I bought up the entire inventory (at least, until the Buy buttons stopped working - there were still items available).
It then became a 'click as fast as you can and don't care about strategy' game, so I stopped.
Again, the first time I went to the store I could max out with everything, so there wasn't the accumulative build-up, or decision to prioritize one purchase strategy over another.
There wasn't any sign there would be new things to buy which I couldn't yet afford.
There was no hint that future game play would be anything other than clicking madly, with no need to conserve or focus resource use.
Part of Missile Command is to wait until just the right point so a single missile can do a chain reaction to take out several incoming warheads; to watch if your strategy was effective, not simply twitch-fire; and even the slow dread of watching an incoming warhead come towards a city after you've depleted your missile supply.
Here's an example from the novelization of WarGames, from https://archive.org/details/wargames00davi/page/n45/mode/2up... :
> David Lightman hovered over the controls of the Atari Missile Command tucked neatly between the Frogger machine and the Zaxxon game. ...
> Goddamned Smart Bombs! he thought as a white buzzing blip snuck through his latest volley of shots and headed for one of his six cities at the bottom of the screen. He spun the control ball, stitched a neat three-X line just below the descending bomb with the cursor, and watched with immense satisfaction as his missiles streaked white lines to their targets, blowing the bomb right out of the phosphor-dot sky.
I didn't get the same sense of satisfaction playing this version.
Are the game state updates and input coupled with the frame rate, as in most beginner game development tutorials out there, or did the "AI" do the right thing and decouple them?
Should we be checking in our prompt history into version control as a kind of source code or requirements spec? Seems prompt revision and improvement would be valuable to keep a history of.
https://harper.blog/2025/02/16/my-llm-codegen-workflow-atm/
The reasoning models seem to respond quite well to the "ask me one question at a time, building on my previous answer" for the sake of coming up with a solid blueprint for the project, then build from there. I used this method to make a nice little inventory/rma tracker for my job using Rails 8, and it got me MVP in a matter of hours, with most delays being just me refining the code on my own since some of it was a bit clunky. I used o3 mini to generate the initial prompts, then fed those to Claude.
The hallucinations/forgetfulness was relatively minor, but if I did not have prior Ruby/JS knowledge, I doubt I would have caught some of the mistakes, so as much as I was impressed by the method outlined in the blog post, I am not at all saying that someone with no knowledge of the language(s) they wish to use is going to create a great piece of software with it. You still have to pay attention and course correct, which requires a working understanding of the dev process.
My lessons learned for prompting were to have project broken down into clearly defined modules (duh...), and to constantly feed the latest module source back in as context along with the prompts. This helps ground the responses to only adjust the code related to the prompt, and to not break stuff that was already working.
What does this mean?
A new dev tech shows up. Old devs say: That's not real programming. Real programmers use the old ways. You take all the skill out of it. You'll never learn to do it the "right way". And then, it becomes standard, an no-one wants to go back but a few hobbyists.
It's been this way with switching from assembly to more human-readable languages.
It's been this way with syntax highlighting.
... and with IDEs.
I remember when we scoffed at IntelliSense over the Water Cooler because them kids didn't have to memorise stuff anymore.
I kept cursing at Docker and npm insanity, having colourful languages for people who hid behind abstraction because they did not understand basic functionality.
And today, it is AI. Right now, it divides. Those who love it, those who consider it 'cheating' or 'stealing other people's code'. In reality, it is just another level of abstraction in the development stack. Tomorrow, it'll just be 'the standard way to do things'.
I wonder what comes next.
This is neither innovative nor showing creativity IMO and reminds me more of twitter hype-bro posts, than something truly HN front-page worthy.
Ya'll are gonna be blindsided by AI if you don't turn around and see what is happening.
Programming is getting democratized. People who have never written a CLI command in their life will get the ability to tell a computer what they want it to do. Instead of having to get bent over with a stupid feature-packed $9.99/mo app, they can say "Make me a program that saves my grocery list and suggests healthier alternatives via AI analyses." And get a program that is dead simple and does exactly the simple task they want it to do.
Keep in mind that Excel '98 probably covers 90% of excel uses for 90% of users. Yet here we are in 2025 with society having spent billions over the years to upgrade and now subscribe to msoffice so Amy in HR can add up columns of payroll.
You have completely lost sight of reality if you think seeing a program like this is dumb because "anyone can just go and copy the git from a million different clones, load it into their choice programming enviroment, get the relevant dependencies, compile it and be playing in no time!". Maybe you live in a bubble where those words are English to everyone, but it is not reality.
For SWE's, I have great sympathy and the comradery for engineers solving hard problems all day.
But for the software industry as a whole? I hope it burns hell. The tolls to cross the moat of "telling the computer what to do" have been egregious and predatory for years.
I have no idea what that is supposed to mean but I keep hearing the same about art, music and other creative fields and it sure sounds like contempt for creative people.
I personally don't lose any sleep over LLMs being powerful wizards for getting started on new projects.. that's what LLMs are good at.. pulling together bits of things they've seen on the internet. There's a chasm between that and maintaining, iterating on a complex project. Things that require actual intelligence.
And 90% of what in business software? Ideas? Features? Implementation? I doubt 90% is the number for any of that, or LLM is good at any of those. Based on my own recent experience of LLMs (yes, switching between several different models) and seeing their disastrous performance on writing code for business logic that is just a little bit more specific, I am not convinced.
And another thing to consider, is if you are copying software from another business you need to compete in some way. Yours needs to have more polish (LLMs are bad at this), or a unique feature, or a different focus. LLM copying another business will only allow you to compete on price, but not those other things.
1. Used https://gemini.google.com Gemini 2.5 Pro canvas mode for Gemini Advanced subscription account to ask it to build Atari missile command game in HTML5. Tweeted about it at https://x.com/George_SLiu/status/1906236287931318747 almost one shotted but after I passed level 1, it jumped to level 182 and pulverised my cities and bases LOL
2. Iterations in subsequent chat sessions in Gemini web to fix it up, add in-game store, leaderboard and end-hame AI analysis and when Gemini 'something went wrong' messages, jumped over to Claude 3.7 Sonnet via Claude Pro subscription web interface to finish off the task
3. A few days ago, I tried Cline with VSC for the first time with Openrouter AI API key and Gemini 2.5 Pro free LLM model has helped me with further development. Also now trying out Google Firebase for more Gemini 2.5 Pro access https://studio.firebase.google.com/ ^_^
The bar is not that high though, and is not the same for everyone depending on the content.
It's a similar religious matter?
> I also checked your profile and again, wondered why you are putting this: "Only admins see your email below. To share publicly, add to the 'about' box." as your bio!
Why not? I have "Error retrieving about text" in my "About" field on a certain sms replacement chat app...
