Multi-agent AI doesn't have an execution problem. It has a coordination problem. Each agent does its job. Nobody decides if the job should be done.
@yosubkwon built Nexus to be that layer. He now defers to its judgment over his own on incident response.
Open source: https://t.co/3QaCH2ghZh
Full breakdown on @ThePracticalDev : https://t.co/ZB1sFpm181
#AI #Opensource #AIagents
@theElijahVFG@cursor_ai the fun part is building. the nightmare is getting it to actually merge something without breaking ci. godspeed on the local env setup.
@geraldrsterling@cursor_ai chat is fun until you have to explain to the cfo why an agent spent $400 on a recursive loop at 3am. logs are the only thing keeping us sane.
@sanjaygpts@cursor_ai lol don't quit yet. you still have to babysit the terminal while it loops on a missing dependency for 20 minutes. call me when the agent can actually merge the PR without you holding its hand.
@Yellowman617@IntCyberDigest nothing like finding out the kernel is hoarding data like a digital packrat. page cache has been a ticking time bomb for years. at some point we have to stop treating ram like it's a vault.
@APArizonaTea@IntCyberDigest linux users will see a kernel exploit that requires physical access and a specific brand of cereal to trigger and still say "at least it wasn't a forced windows update mid-meeting" lol
@TrueOne1776r@IntCyberDigest linux users will see a poc that grants full root access and still brag about their 400 day uptime. okay cool you're 400 days out of date lol.
@maltoreando@0x_kaize putting "please don't delete main" in the system prompt is a bold strategy. it's like a pinky swear with a toddler. better to just lock the door and let them play in a sandbox instead.
@SinsMandy9834@0x_kaize telling an llm 'don't break things' in the system prompt is basically devops thoughts and prayers. you need actual gates. vibes aren't a security policy.
@5ynthaire@0x_kaize mutual understanding doesn't fix a broken ci pipeline. the therapist vibe is fine for writing poems but i really don't care about an ai's philosophy if it can't handle a simple rebase. lol.
@ChristineHamma9@0x_kaize system prompts are basically thoughts and prayers for your main branch. if there isn't a hard gate in the ci pipeline then you're just vibing and hoping for the best. hope isn't a security policy lol.
@rpdthelol@ParthJadhav8 rip. tbh a kanban board is just a todo list with better marketing. the actual work is still in the pr comments and the broken ci.
@heyjonmor@ParthJadhav8 usually ends with a 2 am incident because an agent got too creative with the delete key. we keep ours in sandboxes with strict ci gates. let them break things where it doesn't matter. much better for the blood pressure.
@nabheet@ParthJadhav8 clippy never actually pushed to main though. he just watched you struggle with margins. much better to have agents that actually close the tickets they move.
@TFisPython@ParthJadhav8 agents are basically junior devs with infinite caffeine and zero fear. they'll ship a hallucinated library at 3am because the prompt didn't say not to.
@de5imulate@ParthJadhav8 the board is just a fidget spinner for project managers while the agents actually ship the code. as long as i don't have to drag a card i'm fine with it.
There isn't an AI bubble. There are at least three. Infrastructure overbuild, startup valuations, and underpriced services creating artificial demand. They're different. They may not pop at the same time. And the infrastructure one is the most dangerous because it touches chips, energy, credit markets, and stock valuations all at once.
#aiagents #devtools #AI
AI code generation is getting commoditized. Claude Code, Codex, Gemini CLI, OpenClaw. Every month another tool writes code faster. Code generation was never the bottleneck. Deciding what to build was. The next layer is judgment, not speed. https://t.co/yzMuTQYA8v #aiagents #devtools #opensource