Grok actually shows a number of sources used for an answer. Once I asked it something simple and it apparently scanned 200 different websites. And it was just a short prompt. Now imagine millions of users asking for something multiple times a day.
Agents could work for a long time and burn a lot of tokens if you give them a task that is too hard for them. After enough time, if they don't give up, the slot machine could spit out a working solution.
Personally I find it faster to figure out the hard parts by myself and then give a few smaller tasks to Claude.
That's a lot. On my usual day I burn less than $1 on Opus. I could get beyond $10 only if I have a complex and well-defined problem, which is rare (the second part at least).
Luckily very few people can configure and are interested in local models. But your nearby datacenter running Chinese open-weight models is also good enough.
> Only environments that could be fully solved by at least two human participants (independently) were considered for inclusion in the public, semi-private and fully-private sets.
So far AI doesn't seem even close to replacing senior engieeners. Hell, it can't even replace junior engieeners entirely.
I use AI agents every day at work and I'm happy with that, but it took over two years and billions of dollars in investment to deliver anything useful (Claude Code et al). The current models are amazing, but they still randomly make mistakes that even a junior wouldn't make.
There's another paradigm shift to be made certainly, because currently it feels like we scaled up a bug brain to spit out code. It works great for some problems, but it's not what software developers usually do at work.
I just found out about pi yesterday. It's the only agent that I was able to run on RISC-V. It's quite scary that it runs commands without asking though.
I have Pro Edition and for me Copilot only added two icons. One in Notepad and another one in Paint. I ignore both. There's also the Copilot app that I didn't even know I have installed.
I don't know what happens with Home Edition, but I though the pushback was mainly from Insider Preview?
This. A lot of people on HN acts as you can only write code manually (almost, generators and snippets are allowed, because we are used to them) or vibe coding the whole project through a WhatsApp conversation. As if there was nothing in between and the same approach should work for all kinds of projects.
Personally I use coding agents for boring parts (I really don't enjoy putting the same piece of string to 20 different classes just to register a new component) and they work quite well, I'm going to use them for foreseeable future, because they make coding much more enjoyable for me. On the other hand I don't have an OpenClaw box burning billions of tokens weekly for me, because I usually don't have ideas that could be clearly specified.
That's weird. Back when Java 7 was a new thing, people used Spring (Boot wasn't there yet) even more to compensate for the lack of language features. Also back then most projects still used XML configuration, so you actually write more Java code in modern Spring. Because Spring Boot uses Java configuration classes by default (although you can still use XMLs if you need for some reason).
That's why coding agents usually scans various files to figure out how to work in a particular codebase. I work with very large and old project, and Codex most of time manages to work with our frameworks.
Grok actually shows a number of sources used for an answer. Once I asked it something simple and it apparently scanned 200 different websites. And it was just a short prompt. Now imagine millions of users asking for something multiple times a day.