I gave it the following initial prompt:
> An app where you input a question, then flip some coins to generate an I Ching prediction, and it generates a prediction / fortune for you. Then this combination of results can be fed to Gemini AI to produce a more detailed prediction text.
It generated something that looked fine. When I input a question and press the button nothing happened. After asking it to fix the problem multiple times and having it fail, I looked at the browser console to figure out the errors it was getting. Then I copied those errors and told it to fix them. After a few iterations, it solved every error and would generate a result. It completely forgot the part where you are supposed to flip coins before getting a hexagram to generate a fortune. After a bit of prompting, I was able to get it to display the hexagram and input question. However, sometimes it becomes confused about which hexagram was generated.
Overall, my impression is that these tools are still in the toy novelty stage rather than something you'd want to use for anything important.
Here is a screenshot of the app output for the question: Will Hacker News like my vibe coded oracle? [0] As you can see, it says that the generated hexagram is 24 or 41, but in the fortune text below it says 11.
At first, the product I was using (lovable.dev) seemed to me exactly as you described. I gave it a basic app outline and hit run, and it produced something that superficially looked right but did nothing.
So I asked some other people for advice, and they said you have to hold its hand and go step by step. So I did.
I told it, give me a landing page that matches [product description], but implement nothing else. Then, ok, let's set up auth - add a sign in and sign up dialog. Then, ok, let's create a user account page. Bit by bit.
It succeeded wildly. I was able to build the whole thing in 3 days. I'm not capable of that on my own, it would have taken me 3 weeks. Sometimes the AI got stuck and I had to manually go in and accomplish what I wanted. It took over 100 steps to complete the product, and probably around 10-20 times I had to revert its changes and give it more specific instructions. I had to check its work at every iteration, just like with a junior developer.
But it worked. And it's going to get better. Would I use this for "something important"? Depends how you define that. I used it to build a working product. Would I start letting it modify an existing mature codebase willy-nilly? No, probably not. Would I let it write cryptographic logic or trust that it wrote bulletproof code from a security standpoint in a sensitive context? No.
But for a simple application, it was an incredibly powerful tool. Especially for something that didn't even exist just 2 years ago. Give this a decade and it's going to change all our careers even more than it already has.
I have tried the hand holding approach with Cursor. It doesn't work for me. I have to constantly correct and over correct. Getting auth working sounds insane to me.
This model works really well. For example, simple things like [1], [2], and [3] can apparently be generated with just a couple of prompts.
[1] https://koreanrandom.com/en/games/through-the-space/
[2] https://koreanrandom.com/en/games/tanchiki/
[3] https://koreanrandom.com/en/games/sharik/
According to the author, these were made with Gemini 2.5 Pro without any manual coding by a human.
Cursor isn't as powerful as Gemini in AI Studio because AI Studio gives you full control over the model's settings and how it processes code.
Plus, the massive 1 million token context window is incredibly helpful for working with large codebases. You can use tools like code2prompt and repomix to feed all the necessary context into AI Studio from the clipboard for those projects.
I don't have access to the precise prompt, but I told it something like "only implement basic authentication based on JWT, using just email + password.", then asked it to add a simple registration form, then the password reset flow... step by step, but with little guidance.
At every prompt I review the changes on git, and commit.
```
// It works, but we have no idea how to send an HTTP request in java.
// If you do, send us a nice example to support@errortexts.com.
// You just need to POST JSON with the keys `api_key` and optionally `message`.
```
That's brilliant. But it does make me wonder why the LLM couldn't have provided you a suitable Java code example?
The first version[0] looked good, but when I inspected it I found that it just picked an I Ching prediction at random on the back-end, instead of actually flipping coins.
I updated the prompt to the following:
> Create app where you input a question, then flip coins to generate an I Ching prediction (client-side). First show the standard I Ching prediction and it's hexagram, and then use AI to generate a fortune analysis based on the prediction and your initial question.
And the result was much more laborious[1] of a UI :shrug:
It broke more and more each message. I tried fixing stuff myself but it would mess it up again. Would not recommend anyone to use it.
Now for the non ai part: super cool. I love the nix environment. Its fascinating how they handle the previews for example. I got geekbench up and running an the cpu is a bit worse than an iphone 15 pro max, but it has 32 gigs of ram!
The app prototyping logic in Firebase Studio isn't wired up for Flutter/Dart yet. You can play with Gemini+Dart/Flutter here: https://dartpad.dev/?channel=main.
We're working with the Firebase Studio team to integrate. FWIW, it seems to do fairly well with "Create a chess clock app".
But try this:
Open up a blank Flutter template in Studio: https://studio.firebase.google.com/new/flutter
After everything initializes and your Android and web previews set up, open chat by clicking the little Gemini spark at the bottom of the workspace and then add your prompt.
YMMV, but I got a very basic, but working chess clock in one shot with "Can you replace this sample Flutter project with a fully-functional chess clock that works on Android and on the web?"
To make sure I tried 3 times in a row. Same result.
I feel like in the emperors new clothes story.
Gemini actually dug me out of it, though, with additional prompting (this kind of surprised me; I assumed I’d have to eventually really step in). After all that distraction, though, I had to reset the convo to get it to focus on the files again.
Totally understand that you might be too annoyed at this point, but if you do have a few minutes, it’d be amazing to file a case @ https://firebase.google.com/support/troubleshooter/studio for the team to dig into.
A web GUI for Firestore that lets me work on documents like, idk, any other DBMS GUI would: the ability to select multiple records, and operate on them.
That's literally it. I don't need AI, I don't need dark mode, I don't even need MongoDB compatibility. I just want to select multiple documents with my mouse and do things to them.
Come up with your data model, explain it to the AI, and tell it to give you a CRUD for it.
I'm pretty sure it's perfectly doable to ask it to give you a dynamic crud based on the "shape" of the data in Firestore.
Sadly it's an internal tool for work, so I can't share.
Despite all the recent enshitification, I can't think of any alternative solution that would come even close to what Firebase has to offer. Authentication API is especially hard to beat (cheap and very easy to integrate).
Looking at that, I'm not sure corporate would be down for something like that as a solution for "select multiple documents in a database GUI and operate upon them". Could be wrong though.
The overall chat in the HN conversation has got me thinking, though.
Around 7 years ago in my career, one of my most common actions for one-off scripts was for me to create a WinForms application with, often, a couple text boxes and a "Run" button of some sort.
The text boxes would be the inputs and the run button would ... run. There was also often like a text output or bunch of loglines or something. I wrote almost exclusively in C# at the time, so it was a way to shove a bunch of C# code into place and test it.
I did this for random and arbitrary things I needed to process or solve, a lot like how I used Python or Ruby in the future.
I bet it's actually pretty common for people to need "a script that does a thing", and I think, maybe, that's where a lot of the AI scripting of the most immediate use is going to be. If it can be a familiar interface for people to build (in the past, the IDE) and a familiar or simple place to interact with the generated script (the WinForms + buttons), these programs to generate scripts and do "stuff" could likely spread pretty wide.
I think Jupyter Notebooks are another example of this, another precursor, of sorts?
> You're talking like go to market is rocket science.
In fact, it's not that easy. The second part -- GTM -- is the hardest part; even if you could clone Notion, getting that product in front of customers is the hard part.
https://community.firebasestudio.dev/t/working-fasthtml-a-ne...
I preface because he's been using AI basically every day to build his SaaS.
His comments today:
> also gemini 2.5 is beating claudeBut I can say that as a 20yr+ eng. and having worked with him on a prior startup, I'm super, super impressed how far he got by himself in under 4 months. Basically a full-blown industry specific CRM with Stripe integration, OCR from mobile web app, desktop web app, marketplace. He himself is concerned that it's the end of SaaS as he thinks anyone can build these things now, but I think there's still a bit of a chasm when it comes to some ancillary pieces like databases.
This… warrants some self reflection.
That's kinda telling.
It's like some people are triggered by non-veterans programming with AI. I think everyone understands the downsides. Why get angry about it?
I understand and (mostly) stay out of the arguments between HN users (both "technical" and "non-technical" business types) about AI coding. (And I'm not particularly interested in airing those grievances again in this subthread, well mannered or not.)
But I'm intrigued that the dev mentioned upthread is intentionally hiding his use of LLM coding, instead of joining in with the AI buzzword bandwagon jumping that's going on pretty much everywhere I look in tech, from tiny to behemoth corporations.
Sarcasm is the internet default, though. Perhaps you didn't quite pick up on it that time?
What does that have to do with promoting apps and whether or not it is sensible business decision to do so? You must have accidentally hit the wrong reply button? I think I saw that discussion taking place elsewhere.
But they do lol.
But as people on the internet do find great enjoyment in being funny (for some definition of funny), sometimes their humour might come across as being "angry" if you don't understand the context in which it is taking place.
This is such a bizarre conversation. Yes, people on the internet sometimes get angry and don't go do something else. It happens.
No one is mistaking humor for anger.
I think that's pretty irresponsible - building a product that holds PII without any concern for security. Not to mention that's a lawsuit waiting to happen.
FE: React SPA with Tailwind; I don't think he was knowledgeable enough to pick a UI component library.
BE: Database is Supabase, auth is Firebase. He had a hard time getting the AI to work with Supabase auth so I suggested Firebase instead (Firebase auth is dead simple and lots of docs). This created another challenge which was using the claims from Firebase in Supabase for the row level security. Rest of the backend is running in Replit autoscale deployments. Stripe integration for payments and other integrations for common third party systems in this particular industry. All done via AI and handing it relevant documentation and code samples.
Replit's tooling is interesting because it can run the app and see the errors in the console. It can also take its own screenshots to see if it's getting it right. I watched some of his sessions and the LLM often makes mistakes. He'll try a few passes to get the LLM to fix the mistake, but if it doesn't work, he moves and will come back to it if it is non-breaking.
I talked to him a bit more and he said he thinks his "super power" is managing an offshore team in India for a few years so he's very used to managing features and code quality via written instructions. Where an engineer (like me) might just give up and take it into our own hands when things go wrong, he can just keep chatting with the AI and pointing it to docs and such.
Sorry to say that Firebase Studio did an awful job. It did not successfully build even the first view of the app I asked for. It feels like I'm stepping back to release day of GPT-4.
Am I missing a switch to use the good Gemini 2.5 somewhere? I could tell from their response speed that I was not using a thinking model.
I do most of my vibe coding on my phone, so that's pretty disappointing!
Curious what kind of phone apps you’ve been using to do vibe coding, I suppose it auto deploys stuff also?
That repo automatically deploys to my https://tools.simonwillison.net/ site.
Google and Microsoft both have web IDEs, so it's a natural product evolution for these (though Microsoft's Github is much better positioned with its deep distribution and integrations).
It still feels like these are so far behind Cursor and Zed and the like in terms of getting real work done, but maybe the customer is new or non-engineers.
I haven't looked at startups like Replit. Are they even in this game?
And then there are the end-to-end products like Vercel's V0, Lovable, Bolt, etc. If they can get past the 80% problem, they might be a huge deal.
This is going to be quite the battle.
https://firebase.google.com/docs/gemini-in-firebase#:~:text=...
However, the aistudio updated today is good.
So it seems that the output is so different because the person in charge is different?
From this perspective, there is no hope for Firebase, because the person in charge has poor management logic. even with the best resources in the world, it doesn't help
About the executables, are backend apps running on a temporary VM which can run Postgres etc? Does the Android emulator launch from web?
"Firebase Studio is a powerful visual development platform that lets you build, customize, and deploy full-featured apps with the speed of low-code and the flexibility of Firebase. Whether you're prototyping or scaling to millions of users, Firebase Studio streamlines your workflow with real-time data, seamless backend integration, and built-in hosting."
So, Aider + whatever seems to be working best, seems to be the best strategy that works for me. Aider has this nice feature that I can specify what files are added to the context, so I pay less and I do not confuse AI with stuff unrelated to a given task.
Also, the system prompt encourages it to use Genkit, so it tried to add AI to an email sending function...
You will inevitably run into some doom loops, but we're still at the beginning. These things will probably be super powerful in 2 years time.
Is that possible?
Billings systems seems to be so opaque and complex.
and can it learn my private keys and then expose them to someone else?
I've wanted a small prompt-manager chrome extension for a while.
Was procrastinating.
Was able to build one for myself with Firebase studio in 30 mins.
Here's my PromptPal - built in just 30m (disable ad-blocker to avoid issues - there's some interference for some reason):
https://9000-idx-studio-1744253706406.cluster-fkltigo73ncaix...
No frustration whatsoever
Their prototyper is awesome
And code mode also great
I was able to push to github as well with no problems. And the tool generates nice commits for every single change one makes.
- Add SSR (fail)
- Give the correct command to add SSR (success)
- Add i18n (fail)
- Retries like 6 times, completely messes angular.json file up to no recovery, project doesn't even build
Outputs: It seems I am running circles, I cannot help you with that.
Wow, AI will really replace developers soon. /s