A self-hosted Rails gem was released to attribute LLM spend by feature, user, and tenant. Most of the implementation was generated by LLMs, and it was shipped without thorough reading. Feedback identified problems, and the criticism was accepted as fair. The experience showed that writing code becomes cheaper, while the real work moves to review, accountability, and maintaining correctness. Ongoing cleanup is being done to address generated code issues in the gem. Early adopters are being sought to run real LLM calls in Rails and report what breaks. The project is MIT licensed.
"A month ago I published llm_cost_tracker (a self-hosted Rails gem for attributing LLM spend by feature/user/tenant) - and almost all of it was LLM-generated. I shipped it without really reading it, got called out, and the criticism was fair. This is an honest write-up of what that taught me: writing code got cheap, so the actual work moved to review and ownership. I'm still cleaning up generated code in the gem and looking for early adopters running real LLM calls in Rails to tell me what breaks. MIT."
"A month ago I published llm_cost_tracker (a self-hosted Rails gem for attributing LLM spend by feature/user/tenant) - and almost all of it was LLM-generated. I shipped it without really reading it, got called out, and the criticism was fair. This is an honest write-up of what that taught me: writing code got cheap, so the actual work moved to review and ownership."
"I'm still cleaning up generated code in the gem and looking for early adopters running real LLM calls in Rails to tell me what breaks. MIT."
Read at Rubyflow
Unable to calculate read time
Collection
[
|
...
]