The simpler version I like is when you get stuck, write down why you're stuck and _why what you're doing should've worked_. Usually that's enough to find an incorrect assumption and make traction.
Like you're calling the API and passing the credentials when it expects you to have a cookie set. The API should work because the endpoint is correct and I'm passing user/pass as parameters. You can then validate that the endpoint is correct but when you go to validate the parameters you'll notice its supposed to be a cookie.
If we’re aiming purely at reducing troubleshooting time then asking an LLM at the end of all those lengthy steps like reinstalling environment doesn’t make sense.
Even if they can only one shot solve 10% of the problems it makes sense to do that as a first line of attack since it’s fast and low effort and even the 30seconds it takes can be done concurrently with other troubleshooting.
I’d venture the solution is to keep the LLM as first step but to abandon it very fast if it doesn’t produce immediate results. People get stuck there and make it their main troubleshooting approach and never get to the engage brain part
That's more or less what I've been doing lately. It's great at first stage rubber ducking, and very occasionally my problem is common enough that it already has the solution in the database.
Probably more than half the time I use the LLM to barf up the right keywords then search for the answer myself.
Since my job right now is "try to cover for ten missing employees" it's been vital to staying productive. Along with a general focus on efficient processes, I've managed to get my workflow pretty fast.
The simpler version I like is when you get stuck, write down why you're stuck and _why what you're doing should've worked_. Usually that's enough to find an incorrect assumption and make traction.
Like you're calling the API and passing the credentials when it expects you to have a cookie set. The API should work because the endpoint is correct and I'm passing user/pass as parameters. You can then validate that the endpoint is correct but when you go to validate the parameters you'll notice its supposed to be a cookie.
If we’re aiming purely at reducing troubleshooting time then asking an LLM at the end of all those lengthy steps like reinstalling environment doesn’t make sense.
Even if they can only one shot solve 10% of the problems it makes sense to do that as a first line of attack since it’s fast and low effort and even the 30seconds it takes can be done concurrently with other troubleshooting.
I’d venture the solution is to keep the LLM as first step but to abandon it very fast if it doesn’t produce immediate results. People get stuck there and make it their main troubleshooting approach and never get to the engage brain part
That's more or less what I've been doing lately. It's great at first stage rubber ducking, and very occasionally my problem is common enough that it already has the solution in the database.
Probably more than half the time I use the LLM to barf up the right keywords then search for the answer myself.
Since my job right now is "try to cover for ten missing employees" it's been vital to staying productive. Along with a general focus on efficient processes, I've managed to get my workflow pretty fast.
Interesting to note that "staring at the screen confused" and "systematically working through the problem" look almost identical from the outside.