Having worked on very old government systems for many years as a software developer I can say I completely agree with @SecDuffy.
They really do need to fund and update those systems. Badly.
I stopped using routines and doing complex tasks because of how many tokens it was eating.
I really hope they fix it. I still cant trust cursor agents until there’s some improvement.
@Zitronom I’ve found a good skill in the future is going to be figuring out the line between what the agent should do and what a script or some deterministic program can handle.
What repetitive tasks am I making an agent do that a script should is a daily question for me.
@building_risk It’s a little of both in my case.
It seems to be for me that they like to send off cloud agents to do stuff without fully grasping what needs to happen really eagerly. I have to stop them and say hold off until we’ve get this right, even with standing rules in place.
She was such a bad candidate. I can’t believe I crawled over broken glass to vote yesterday with 3 kids under 3, had to take off work, have been sick, and this is the outcome.
We’re all pretty rightfully upset at the establishment. Literally if you picked some random person off the beach and ran them instead they’d have had just as much a chance of winning. Absurd.
Just as a general tip, cursor cloud agents use way more than a local grok CLI build. I ran out of tokens on Grok Bot and then topped off like $20 to see if I could finish a small task. The cursor agent ate the $20 in less than 5 minutes and didn’t finish the task. It was a small coding task, targeted bugfix like I do all the time. Had to stop and pick back up. Not sure if that eats tokens but it was definitely eye opening.
So, I’m sticking to the CLI for coding for now.
@AI_EmeraldApple Exactly. That post was like a live test on the Dunning Kruger effect with how many said “I learned this in grade school, it’s 9” and didn’t stop to think that maybe there’s a reason those of us with degrees in this immediately come up with 1.