hi @X! I'm Erv, 23, AI engineer.
co-founder and CTO of GoFound, launching soon.
looking to connect with people into:
- AI engineering and LLM apps
- system design and backend
- SaaS and startups
say hi and I'll follow back
hi @X! I'm Erv, 23, AI engineer.
co-founder and CTO of GoFound, launching soon.
looking to connect with people into:
- AI engineering and LLM apps
- system design and backend
- SaaS and startups
say hi and I'll follow back
When I say a screen looks wrong and my AI agent says the code is correct, the screen wins. The agent has to trace the full layout chain before it can tell me nothing is broken.
My local AI box once refused to boot: fans at full speed, no picture. I was sure the GPU was dead, but it was a RAM stick that wasn't seated right. Nothing was broken. Check the board's debug lights before you assume a part is dead.
My local AI box was pulling a model at 200 KB/s and kept dropping off the network. The cause was one setting: Wi-Fi adapter power saving. With that off, the same download ran at 26 MB/s. What boring setting cost you the most time?
I let my AI agent pick the target disk for an OS reinstall. The agent chose by size, which turned out to be the old hard drive, and the SSD sat empty. The PC was painfully slow until I asked why. These days it shows me the disk list with types before it wipes anything.
Small lesson from working with an AI agent: during cleanup it stopped a dev server without asking. New rule in its instructions: check what is running on a port and ask me before stopping anything. The full list so far is in the image. What is on yours?
I don't learn my codebase file by file. I learned one map of the app once: every part and how they connect. Every change gets explained against that map.
@YilmazcimUyar Hi Yılmaz. I build software with AI agents and post about the process around it: the checks, the reviews, what breaks. Scripts that do chores is my favourite part of your list. What chore did you automate last?
@AbdulSonaike Most of mine start there. When I look at the fixes I ask my agent for, they're usually gaps in my brief, not bugs in the code. The rule that helped: before it plans anything, it has to ask me questions first. Does the logic checker also catch missing rules, or only clashing ones?
@unicorn_oz No. What replaced reading for me: tests, a second model's review on anything risky, and screenshots of the feature working in the real app. The part I still own is the list of things that need my yes: money, logins, deleting data. What do your agents have to ask you about?
I don't read the code my AI agents write. This is what every change has to pass instead. The risky ones (money, logins, deleting data) get a second model's review and still wait for my yes. What would you add?
@thtbee_ Frontend testing is where browser agents earn their keep for me: after every change an agent drives the real app and attaches screenshots as proof. The rule that matters most is that it checks the page really changed after each step before moving on.