I don't understand how random thoughts on X are front-page news on Hacker News.
If some tech CEO makes a major announcement on X, it's newsworthy and belongs here. Anything else that's actual news is also fair game ... but all other X posts do not belong here!
Karpathy is a notable researcher and broader AI leader. Among many, many other things, he invented the term "vibecoding". He also recently posted his autoresearcher project, which is using a swarm of agents to optimize the LLM training and recently produced a training process that is the fastest to achieve GPT-2-level performance using a very small model.
criteria for hitting HN frontpage is generally whether it is interesting to people. That thought is likely the same thought a bunch of us are having at the moment.
It easily reaches the front-page for people with a following. I don't think many votes are necessary to get to the front page. And when there's some critical insight or leak.
Aside from that I've seen few posts on X that didn't follow the pattern, and were short lived at the top.
> On-Topic: Anything that good hackers would find interesting. That includes more than hacking and startups. If you had to reduce it to a sentence, the answer might be: anything that gratifies one's intellectual curiosity.
You can just downvote anything you don't like and move along.
For now, there's simply no convention or established pattern for this yet. Given that, the speed of change, and the lower barrier of creation now, it's almost always better to create your own for this.
Using someone else’s software in the exploration phase is like chewing someone else’s gum.
Yeah I vibe coded a simple app that takes an org-mode file, renders it as a kanban board, and lets me spin up agents for each task with the prompt in the body in a named tmux session. The frontend gets updated via Claude code hooks when an agent is idle.
I think the key is to combine human and agent task tracking in one pane of glass.
To me this reads like trying to fit a solution we know to a problem we haven't yet defined.
The problem is more around ops / visibility / delegation / orchestration of agents, but the solution is being misslabelled as "IDE" which I feel like is the wrong analogy although the right "in-between" step towards what the next thing will be.
I've been working on something along those lines (multi-agent orchestration IDE) for the last few months as a personal project.
There are also a lot of projects out there approaching this from many different angles.
Curious what features people would like to see in an Agentic IDE? Would you like to instruct multiple agents in real time (like vibe coding on steroids) or dispatch autonomous agents to solve a long-running task? Something else?
That gives you a chatbox tacked onto an IDE, not exactly an agentic command center. Cursor gets close. But it’s hard to work on multiple things at once, or across multiple codebases.
I've been working on re-imagining the useful parts of Antigravity (Agent Manager) into an orchestrator that is tightly coupled with an LLM-optimized spec: https://thinkwright.ai/plexus
I really feel this. Every implementation so far hasn't felt like it reduced the contextual load involved for dealing with multiple agents. Tmux/Cmux is great, but whoever figures this out will probably make it big.
It works well enough for my use cases so I don't know what these folks are looking for. I have it configured to run everything in WSL sandbox so the blast radius is limited to the VM w/ the code.
I don't understand how random thoughts on X are front-page news on Hacker News.
If some tech CEO makes a major announcement on X, it's newsworthy and belongs here. Anything else that's actual news is also fair game ... but all other X posts do not belong here!
Karpathy is a notable researcher and broader AI leader. Among many, many other things, he invented the term "vibecoding". He also recently posted his autoresearcher project, which is using a swarm of agents to optimize the LLM training and recently produced a training process that is the fastest to achieve GPT-2-level performance using a very small model.
criteria for hitting HN frontpage is generally whether it is interesting to people. That thought is likely the same thought a bunch of us are having at the moment.
It easily reaches the front-page for people with a following. I don't think many votes are necessary to get to the front page. And when there's some critical insight or leak.
Aside from that I've seen few posts on X that didn't follow the pattern, and were short lived at the top.
> On-Topic: Anything that good hackers would find interesting. That includes more than hacking and startups. If you had to reduce it to a sentence, the answer might be: anything that gratifies one's intellectual curiosity.
You can just downvote anything you don't like and move along.
For now, there's simply no convention or established pattern for this yet. Given that, the speed of change, and the lower barrier of creation now, it's almost always better to create your own for this.
Using someone else’s software in the exploration phase is like chewing someone else’s gum.
Yeah I vibe coded a simple app that takes an org-mode file, renders it as a kanban board, and lets me spin up agents for each task with the prompt in the body in a named tmux session. The frontend gets updated via Claude code hooks when an agent is idle.
I think the key is to combine human and agent task tracking in one pane of glass.
To me this reads like trying to fit a solution we know to a problem we haven't yet defined.
The problem is more around ops / visibility / delegation / orchestration of agents, but the solution is being misslabelled as "IDE" which I feel like is the wrong analogy although the right "in-between" step towards what the next thing will be.
I've been working on something along those lines (multi-agent orchestration IDE) for the last few months as a personal project.
There are also a lot of projects out there approaching this from many different angles.
Curious what features people would like to see in an Agentic IDE? Would you like to instruct multiple agents in real time (like vibe coding on steroids) or dispatch autonomous agents to solve a long-running task? Something else?
VSCode + any LLM plugin solves all the problems for me. Keep it simple.
That gives you a chatbox tacked onto an IDE, not exactly an agentic command center. Cursor gets close. But it’s hard to work on multiple things at once, or across multiple codebases.
I've been working on re-imagining the useful parts of Antigravity (Agent Manager) into an orchestrator that is tightly coupled with an LLM-optimized spec: https://thinkwright.ai/plexus
Early days and would appreciate any feedback
I really feel this. Every implementation so far hasn't felt like it reduced the contextual load involved for dealing with multiple agents. Tmux/Cmux is great, but whoever figures this out will probably make it big.
The VSCode forks all do too much, Nimbalyst is built from scratch to be a proper agent manager. https://nimbalyst.com/
Intent from augmentcode is trying to be this https://www.augmentcode.com/product/intent
who the hell made that demo video. i want to quickly see how it works not unskippable video of some person blabbering.
they had my attention. now they lost it.
I should probably try cmux+worktrunk again, but agent-of-empires works pretty good so far.
Antigravity is getting there
It works well enough for my use cases so I don't know what these folks are looking for. I have it configured to run everything in WSL sandbox so the blast radius is limited to the VM w/ the code.
maybe he can vibecode one himself. i know i did.