Gerald
92 posts

Gerald
@gerald_was_here
The window to build quickly without being a dev is open. AI hasn't killed SaaS. Yet. Sprinting through this window before it closes
Katılım Kasım 2025
38 Takip Edilen6 Takipçiler

@bil0090 for anyone trying this: pin claude to read-only first, let it map the disk, then let it delete. last time i ran a similar audit it nuked an old xcode derived data dir i was actively using
English

Claude just saved me ~$5K in 1h 🤯
So, I have a 1TB MacBook, I had 100% used, only 2GB left
Was about to buy a new MacBook, but...
I went ahead & just told Claude to audit my MacBook & free up some storage, without deleting anything important
I had to accept everything too, just to be safe
After about 1h, it literally freed up 508GB, which is nuts
Nearly half my storage is free
It saved me from spending ~$5K on a new 2TB MacBook
It also solved a memory issue I was having & made my MacBook faster, like new
Next: I will be trying the security audit idea that @levelsio shared 🔥
Thanks @AnthropicAI, will reinvest this money in Claude again :)


English

Gave Gemini 3.1 Pro global permissions in Antigravity and a large .MD prompt file with instructions to spawn subagents for:
- adversarial review
- detect drift from my original prompt and intent
- check architecture and efficiency
Then I literally took a nap.
The goal was to fix large issues in my programmatic WhatsApp style video generator.
The result is stunning. This is just one of the test files it made on its own. It generated 6. All flawless.
If you aren't already, GIVE YOUR AI BIGGER TASKS
English

@Saanvi_dhillon Gemini by far. $10 plan and the usage limits are practically infinity (?)
Been using nonstop for 4 hours and apparently at 1% usage for 5 hours lol

English

My first product is a budget bot that I'm launching in Peru/LATAM.
Why not the US? Because it's overcrowded and there's waaay too much competition.
In Peru it's an open field, plus I've spent a large part of the last 4 years here (I'm in Lima now).
So why not? Lower profits per user, sure, but volume matters too.
Not sure how it'll go, but the experience I'm getting while building this is hugely valuable.
Win win.
English

Been building a WhatsApp video generator as distribution for a small product I'm making when I realized...
I didn't check if this was already available 🤦♂️
Upon checking, there are things available but they're not very good, so I'm going to continue building my own.
If you're out there building tools, research if what you're building is already available!
Things really can (sometimes) be that easy
English

@Miguel07Code @HyperFrames_ Thank you I appreciate that Miguel 💯
English

@gerald_was_here @HyperFrames_ You can forward me here any render issue that wasn't caught by hyperframes lint/validate before rendering the video or open an issue in the hyperframes repo attaching a repo reproducing the issue
English

Rebuilt my WhatsApp video generator using @HyperFrames_ instead of Remotion today.
Hyperframes is meant to be easier to work with for AI agents. So far so good. Some issues with the render, but we'll see if troubleshooting the bugs becomes easier.
With Remotion it was a pain...
English

@EternalBahamut @naturedotcom Yeah that's the only issue, it does seem to burn an awful lot of tokens. If I upgrade to the $100 plan on anything it'll probably be codex though. At least as of now.
English

@gerald_was_here @naturedotcom Not sure if codex has better limits, the weekly consumption at least to me is awful.
Althought it might be, as i feel is also faster, more efficient and over all more usable.
English

With code, I generally create a comprehensive test suite to test core functionality and edge cases
Then I tell Claude/ Codex to build a feature, run the tests, fix issues, repeat
But with programmatic video generation, like with remotion (or now hyperframes), it's harder because AI still can't accurately spot visual bugs.
Even building a test suite that analyzes frames around transitions and key sensitive animations doesn't work reliably and doesn't catch big visual issues.
Tried a couple MCP servers where the AI "watches" the video. No luck. Tried having GPT and Gemini analyze the videos. No luck either.
Has anyone solved this?
English












