Imagine you have thousands of docs like a bunch of catalogs or product pages. A vector search takes milliseconds and costs next to nothing. Having an agent potentially check over those thousand files for a match could take a lot of tokens and time and the larger it grows the worse the performance and token cost
I've worked with 3 founders in the last 6 months who made some pretty damn good vibe coded apps. These 5 things would've saved them a lot of time and money https://t.co/lV6LDesRF3
So you guys think normal people are really building personal software like that eh?
I truly wonder whether X users have any friends or family sometimes 🙏
@ManiishSharma_ I could’ve made this list a lot longer honestly but I didn’t add those! No logs or rate limits was killing me in the last project 🤦♂️- I need a part 2 now
@jacobAlswin Oddly enough, I've spoken to more people in the last couple months who want to learn to code than in the previous 12. I don't want to over-extrapolate tho cuz I'm certainly not coding shit anymore except for funsies
Worked with 3 founders this year to help save and/or ship their vibe coded apps.
Way more fun than I imagined and the challenges have been interesting:
- One was not using Git and instead allowing agents to write directly on the box
- Another skipped using frameworks and had a MASSIVE html/css/js file that did in fact "work"
- Agents that basically relied on regex to determine intent
- No observability or tests in any of them
What's worse is that there are shady contractors who take advantage of these people and charge outrageous prices to further slop-ify the codebases.
I'll be writing a lot more about how founders can avoid these pitfalls going forward.
All my AI predictions were wrong...
Vibe code clean up specialist is a real thing. I somehow became one.
MCP is honestly really useful. I dismissed this way too early.
Turns out you can 10x. I'm honestly close.
Harness engineering is difficult and valuable.
The models did get cheaper.
Follow me for more failed predictions.