When you are reviewing code, if you don't have any system to reliably evaluate your code complexity, then you can't run autonomous agents. That's where the real bottleneck lies and that's what I'm bulding at Kurtel
If we could make an actual benchmark on hallucination rates (one that is not already infected), I think Sonnet-5 despite what benchmarks say, would be rank much lower than it currently is...
Didn’t know that. I never actually used the desktop app, I will give it a try someday but I think it’s more like a gadget or a way for non technical people to code. I don’t know if you have any reasons to switch from the CLI to the desktop other than you are not comfortable with the CLI.
The biggest bottleneck in cloud agents is that there is no way to make sure the agent did the right thing. We can't evaluate those agents, that's why current cloud agents adoption is not yet adopted by most companies but I think it will in the future as it is one of the biggest productivity hack for any companies and any jobs (when it works).
kurtel open-source library just crossed the 2k downloads ! It's free to use and can save you up to 34% in tokens cost when using tools like claude code. Here is the link : https://t.co/SQY80F78il