This type of “first shot success” is an easy test to run but not really indicative of how useful the tools are in practice. It’s more important to test how they navigate a large codebase, make reliable changes in line with existing code, not introduce random regressions, be fast, etc.
Microservices don't make it any better. They tend to have a lot of implicit contracts that people don't necessarily express well through the API. All it does is give you a boundary that is fuzzy depending on the protocol used to communicate between the services.
Yes, I’m beginning to get tired of these advert-blog posts, particularly those that don’t significantly advance the narrative beyond what’s already been discussed.
Both apps involved email for magic link features, but then both apps genuinely insist on harvesting your email for instantdb before you can view them?. Gotta love living on the post-shame internet.
For what it's worth, we added auth so there was some way for users to edit songs. LLMs still struggle with more complicated permissions for guests. This is something we'll work on though.
I am just starting to feel that GPT-5 is more hype.
Just a day before GPT-5 launch, I made a video about making a tool with agents, Claude Sonnet 4 and GitHub Copilot.
There was so much hype on the launch day - on how good GPT-5 is, how it gets the code right the first time, and how little direction it needs.
So I was compelled to try it again with GPT-5 preview available in Copilot.
And for some reason, it struggled to align with directions.
a. It would make its own decisions, misaligned with what I mentioned. I had to explicitly say "DO NOT DO this....". ( explicit instructions were followed.)
b. It did not complete tasks and moved forward. This is the same style I used with both GPT-4o and Claude Sonnet 4.
Also, it might be good at coding, but it does feel geeky - need to try more.
This type of “first shot success” is an easy test to run but not really indicative of how useful the tools are in practice. It’s more important to test how they navigate a large codebase, make reliable changes in line with existing code, not introduce random regressions, be fast, etc.
Its funny you say that because everyone just started moving back to monoliths, and vibe coding would fit perfectly with microservices lol
Microservices don't make it any better. They tend to have a lot of implicit contracts that people don't necessarily express well through the API. All it does is give you a boundary that is fuzzy depending on the protocol used to communicate between the services.
It's an advertisement.
Yes, I’m beginning to get tired of these advert-blog posts, particularly those that don’t significantly advance the narrative beyond what’s already been discussed.
Vibe-coding an email collection service is wild, intentional or not.
Both apps involved email for magic link features, but then both apps genuinely insist on harvesting your email for instantdb before you can view them?. Gotta love living on the post-shame internet.
For what it's worth, we added auth so there was some way for users to edit songs. LLMs still struggle with more complicated permissions for guests. This is something we'll work on though.
Thanks for temp mail services.
I am just starting to feel that GPT-5 is more hype.
Just a day before GPT-5 launch, I made a video about making a tool with agents, Claude Sonnet 4 and GitHub Copilot.
There was so much hype on the launch day - on how good GPT-5 is, how it gets the code right the first time, and how little direction it needs.
So I was compelled to try it again with GPT-5 preview available in Copilot.
And for some reason, it struggled to align with directions.
a. It would make its own decisions, misaligned with what I mentioned. I had to explicitly say "DO NOT DO this....". ( explicit instructions were followed.)
b. It did not complete tasks and moved forward. This is the same style I used with both GPT-4o and Claude Sonnet 4.
Also, it might be good at coding, but it does feel geeky - need to try more.