Test driving a different way of working with an ai agent. This was an emotional roller coaster. No wonder itβs exhausting. Started great, a plan and change log, giving more to the agent. Eventually agent is like: watch me overcomplicate this. New project, no existing code.
I don't know how they do it; haven't had to look. The difference in operation is showing that Claude/Codex grep and sed like a boss, that's no different than a senior engineer with Vim. Cursor intelligently pulls and adjusts context. Usually faster. Still exploring Claude as well
Cursor is still incredible. I keep changing AI agents, workflows, etc. When I come back to Cursor, it's refreshing. A mistake in instructions can be corrected, unlike other tools. I am digging Claude, but I can't get these results. Codex is a joke in comparison.
It's a simple example of why senior engineers don't always like AI up front. It's not that we don't want to give up on code... It's too painful to watch. Even if it were correct, the suggested solution was to manually build the same payload. But it was also confidently wrong.
Simple example of AI making odd choices. The price_json is `{"ticker": "GOOGL", "price": 380.4849853515625}`. It didn't fix it. Suggested a dumber solution in the same vein. If dict vs. json is the issue, why not JSON -> dict? It's not the actual issue, but if it were...
After multiple instances of the same error, AI told me, "You're absolutely right."... It reread the docs and found the wrong key. The tutorial I was following had it correct, but this bot autocompleted the wrong thing while I was typing.
Anyone else with teenage boys thought about renaming them to tactical error? Currently spawn in my phone (I was spawn point in his), might be time to rename to Tactical Error
It did a good job, with specific instructions. I told it how the branches came to be and what I wanted pulled into my new branch, and told it not to touch my reference branch. I still had to know the code very well and tweak a few things, but overall it went well.
Wow. I didn't think this experiment would go as well as it did. I had multiple branches, and I had a large PR to split. After I split PR 1 and 2, I had to get through the bot fights; it changed a bit. I wasn't looking forward to the rebase of PR 2, so I asked Cursor to do it.
What I hate most about MN. Allergy season, conveniently allergic to most MN foliage. Sneezing after a heavy core workout might kill me. Definitely donβt stand too close to walls. Oyβ¦ the things we suffer for a gorgeous summer
Talking to my kid about homing pigeons, I asked if he knew what they were.
Kid: Yeah, they're like pidgeon besties
Me: WTF
Kid: You know, your homies
Me: <laughed so hard>, just dead, pulled myself together. HomING pigeons, not homie pigeons
The about page content (currently the hero) lives only in your head for now β whenever you create src/pages/about.tsx the hero copy is there to move into it.
bot: user used the wrong parameter, let me make their script 100 lines long and bulletproof
me: reject, again.
bot: massive scripts
me: Reject. Asked why do you keep refactoring everything?
bot: oh you have the wrong param, I'll stop
I am a hungry learner and thrive on change. This is the first time in history(now-2yrs) that I felt this way. Before, if you had the drive to learn, you were a top performer. Now, it's survival. Managers expect 10x out of the box; we have to step back and learn first to get that.
I've never felt this much behind as a programmer. The profession is being dramatically refactored as the bits contributed by the programmer are increasingly sparse and between. I have a sense that I could be 10X more powerful if I just properly string together what has become available over the last ~year and a failure to claim the boost feels decidedly like skill issue. There's a new programmable layer of abstraction to master (in addition to the usual layers below) involving agents, subagents, their prompts, contexts, memory, modes, permissions, tools, plugins, skills, hooks, MCP, LSP, slash commands, workflows, IDE integrations, and a need to build an all-encompassing mental model for strengths and pitfalls of fundamentally stochastic, fallible, unintelligible and changing entities suddenly intermingled with what used to be good old fashioned engineering. Clearly some powerful alien tool was handed around except it comes with no manual and everyone has to figure out how to hold it and operate it, while the resulting magnitude 9 earthquake is rocking the profession. Roll up your sleeves to not fall behind.