Quesiton for y'all engineers—when you're using ope...
# general
c
Quesiton for y'all engineers—when you're using open source projects, do you normally go to the docs/site or to the GH repo and look for examples in the code?
n
Both, but usually start with the docs.
🙌 2
p
I usually start with docs for any discussion of high level architecture and then straight to the test suite and examples for usage.
🙌 1
b
Just let AI explain it to me 🙂
🧐 1
Kind of joking, but I can open Cline, and ask it how to build a Java (or whatever) app using LittleHorse. Docs are for reading on deep topics... Examples explain the patterns to AI
IMO, every startup should be publishing the prompts for successfully building an app with your product.
"I'd like to create two spring boot applications that are connected by LittleHorse"... If you put that in Cline, what keeps me from building, testing, deploying, or making this production ready?
What goes wrong if I type into v0 "I'd like to build a NextJS app that uses Nile for storage and Nile Auth for multi-tenant authentication"?
g
@Bill Tarr this is really great advice. Thank you!
Anecdotally, the thing that goes wrong is that we made some breaking changes between two versions of our SDK, and I can’t stop LLMs from mixing them up :(
Colt is smarter than me and is religious about compatibility, so I hope his vibe coders are happier than mine.
c
I almost banned AI on my engineering team.
The results of AI coding were…spotty at best
g
I learned from bitter experience that its best to let engineers work however they are comfortable. If you push AI, they hate it. If you ban AI, they hate it. I just them accountable to the results.
👍 3
c
Docs are for reading on deep topics... Examples explain the patterns to AI
Do you think AI would end up being trained on a github "examples" repo?
g
It probably already was
c
Good to know...it's eassier to dump code into GH than to write good docs-friendly examples.
g
100% agree
c
😞
b
lol, I asked Cline to code an LittleHorse example with 2 Spring Boot apps communicating when we talked. It found some repos to add to the Maven file, but didn't actually do anything with them... basically said "usually you would connect via messaging, but I just used HTTP". AI throwing an attitude again.
g
hehehe
@Colt McNealy you should do a "vibe code littlehorse workflow" competition on LinkedIn. Winner is whoever uses fewest tokens to get an LLM of their choice to generate a working littlehorse flow.
First prize: Pony ride? 🤣
b
lol
going down this road further - check this proposal out https://llmstxt.org/
g
everyone needs one of those 🙂 https://www.thenile.dev/docs/llms.txt
b
nice 🙂
g
We should probably curate it better
b
did AI generate it?
g
Indeed
b
using AI to tell AI how to understand stuff
g
I figured they should know 😂
b
nope 🤣
but it's efficient anyway
g
Yeah. LLMs are about as introspective as humans... which is very little.
b
lol, exactly
c
Hah! You have no faith in our species Gwen