Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Eight more months of agents

crawshaw.io

Post reply on HN
Print view View original

31–40 of 250 posts

Re: Eight more months of agents

#31

Post by rahimnathwani »

emmawirt wrote: post #16

Curious what you mean by "agent harness" here... are you distinguishing between true autonomous agents (model decides next step) vs workflows that use LLMs at specific nodes? I've found the latter dramatically more reliable for anything beyond prototyping, which makes me wonder if the "model improvement" is partly better prompting and scaffolding.

An agent harness is what enables the user to seamlessly interact with both a model and tool calls. Claude Code is an agent harness.

  ┌────────────────────────────┐
  │           User             │
  └──────────────┬─────────────┘
                 │
                 ▼
  ┌────────────────────────────┐
  │       Agent Harness        │
  │   (software interface)     │
  └──────┬──────────────┬──────┘
         │              │
         ▼              ▼
  ┌────────────┐ ┌────────────┐
  │   Models   │ │   Tools    │
  └────────────┘ └────────────┘
Here's an example of a harness with less code: https://github.com/badlogic/pi-mono/blob/fdcd9ab783104285764...

Re: Eight more months of agents

#32

Post by NitpickLawyer »

almostdeadguy wrote: post #2

In the past couple days I've become less skeptical of the capabilities of LLMs and now more alarmed by them, contra the author. I think if we as a society continue to accept the development of LLMs and the control of them by the major AI companies there will be massively negative repercussions. And I don't mean repercussions like "a rogue AI will destroy humanity" per se, but these things will potentially cause massi…

There are some good things here:

First, we currently have 4 frontier labs, and a bunch of 2nd tier ones following. The fact that we don't have just oAI or just Anthropic or just Google is good in the general sense, I would say. The 4 labs racing each other and trading SotA status for ~a few weeks is good for the end consumer. They keep each other honest and keep the prices down. Imagine if Anthropic could charge 60$ /MTok or oAI could charge 120$ /MTok for their gpt4 style models. They can't in good part because of the competition.

Second, there's a bunch of labs / companies that have released and are continuing to release open mdoels. That's as close to "intelligence on tap" as you can get. And those models are ~6-12 months behind the SotA models, depending on your usecase. Even though the labs have largely different incentives to do so, a lot of them are still releasing open models. Hopefully that continues to hold. So not all control will be in the hands of big tech, even if the "best" will still be theirs. At some point "good enough" is fine.

There's also the thing about geopolitics being involved in this. So far we've seen the EU jumping the gun on regulation, and we're kinda sorta paying for it. Everyone is still confused about what can or cannot be done in the EU. The US seems to be waiting to see what happens, and China will do whatever they do. The worst thing that can happen is that at some point the big players (Anthropic is the main driver) push for regulatory capture. That would really suck. Thankfully atm there's this lingering thinking that "if we do it, the others won't so we'll be on the back foot". Hopefully this holds, at least until the "good enough" from above is out :)

Re: Eight more months of agents

#33

Post by codebolt »

63 wrote: post #28

I have no problem with experienced senior devs using agents to write good code faster. What I have a problem with is inexperienced "vibecoders" who don't care to learn and instead use agents to write awful buggy code that will make the product harder to build on even for the agents. It used to be that lack of a basic understanding of the system was a barrier for people, but now it's not, so we're flooded with code wr…

Where are you encountering all this slop code? At my work we use LLMs heavily and I don't see this issue. Maybe I'm just lucky that my colleagues all have Uni degrees in CS and at least a few years experience.

Re: Eight more months of agents

#34

Post by post-it »

> Agent harnesses have not improved much since then. There are things Sketch could do well six months ago that the most popular agents cannot do today.

I think this is a neglected area that will see a lot of development in the near future. I think that even if development on AI models stopped today - if no new model was ever trained again - there are still decades of innovation ahead of us in harnessing the models we already have.

Consider ChatGPT: the first release relied entirely on its training data to answer questions. Today, it typically does a few Google searches and summarizes the results. The model has improved, but so has the way we use it.

Re: Eight more months of agents

#35

Post by post-it »

codebolt wrote: post #33
63 wrote: post #28

I have no problem with experienced senior devs using agents to write good code faster. What I have a problem with is inexperienced "vibecoders" who don't care to learn and instead use agents to write awful buggy code that will make the product harder to build on even for the agents. It used to be that lack of a basic understanding of the system was a barrier for people, but now it's not, so we're flooded with code wr…

Where are you encountering all this slop code? At my work we use LLMs heavily and I don't see this issue. Maybe I'm just lucky that my colleagues all have Uni degrees in CS and at least a few years experience.

> Maybe I'm just lucky that my colleagues all have Uni degrees in CS and at least a few years experience.

That's why. I was using Claude the other day to greenfield a side project and it wanted to do some important logic on the frontend that would have allowed unauthenticated users to write into my database.

It was easy to spot for me, because I've been writing software for years, and it only took a single prompt to fix. But a vibe coder wouldn't have caught it and hackers would've pwned their webapp.

Re: Eight more months of agents

#36

Post by Applejinx »

gadflyinyoureye wrote: post #22
almostdeadguy wrote: post #2

In the past couple days I've become less skeptical of the capabilities of LLMs and now more alarmed by them, contra the author. I think if we as a society continue to accept the development of LLMs and the control of them by the major AI companies there will be massively negative repercussions. And I don't mean repercussions like "a rogue AI will destroy humanity" per se, but these things will potentially cause massi…

I see them as powerful and dangerous. The goal for decades now is to reduce the human population to 500 million. All human technology was pushed to this end, covertly. If we suddenly have a technology that renders white collar workers useless, we will get to that number faster than expected.

I don't believe that is true, but if it WAS true that human technology was covertly pushed to this end: there are people out there who are demanding that this technology come up with social manipulations (using language) to reduce the human population to a SPECIFIC 500 million.

Or less.

And I don't think it's collar color they're going to be checking against.

So I guess I'm saying I agree that this is powerful and dangerous. These are language models, so they're more effective against humans and their languages. And self-preservation, empathy, humanity do not play a role as there is nobody in there to be offended at the notion of intentionally killing more than 9/10 of humanity… for some definitions of humanity, ones I'm sympathetic to.

Re: Eight more months of agents

#37

Post by Herring »

gradus_ad wrote: post #19
Herring wrote: post #18

> In 2000, less than one percent lived on farms and 1% of workers are in agriculture. That was a net benefit to the world, that we all don't have to work to eat. The jury's still out on that one, because climate change is an existential risk.

Existential? Maybe to beachfront property owners

Everybody gangsta until the permafrost starts leaking massive amounts of methane.

Re: Eight more months of agents

#38

Post by bowsamic »

codebolt wrote: post #33
63 wrote: post #28

I have no problem with experienced senior devs using agents to write good code faster. What I have a problem with is inexperienced "vibecoders" who don't care to learn and instead use agents to write awful buggy code that will make the product harder to build on even for the agents. It used to be that lack of a basic understanding of the system was a barrier for people, but now it's not, so we're flooded with code wr…

Where are you encountering all this slop code? At my work we use LLMs heavily and I don't see this issue. Maybe I'm just lucky that my colleagues all have Uni degrees in CS and at least a few years experience.

The issue isn't when the programmers start using it. It's when the project managers start using it and think that they're producing something similar to the programmers

Re: Eight more months of agents

#39

Post by kevmo314 »

post-it wrote: post #34

> Agent harnesses have not improved much since then. There are things Sketch could do well six months ago that the most popular agents cannot do today. I think this is a neglected area that will see a lot of development in the near future. I think that even if development on AI models stopped today - if no new model was ever trained again - there are still decades of innovation ahead of us in harnessing the models we…

Really? I hardly think it's neglected. The Claude Code harness is the only reason I come back to it. I've tried Claude via OpenCode or others and it doesn't work as well for me. If anything, I would even argue that prior to 4.6, the main reason Opus 4.5 felt like it improved over months was the harness.

Re: Eight more months of agents

#40

Post by giancarlostoro »

post-it wrote: post #35
codebolt wrote: post #33

Earlier quoted context omitted.

Where are you encountering all this slop code? At my work we use LLMs heavily and I don't see this issue. Maybe I'm just lucky that my colleagues all have Uni degrees in CS and at least a few years experience.

> Maybe I'm just lucky that my colleagues all have Uni degrees in CS and at least a few years experience. That's why. I was using Claude the other day to greenfield a side project and it wanted to do some important logic on the frontend that would have allowed unauthenticated users to write into my database. It was easy to spot for me , because I've been writing software for years, and it only took a single prompt to…

You can also ask Claude to review all the code for security issues and code smells, you'd be surprised what it finds. We all write insecure code in our first pass through if we're too focused on getting the proof of concept worked out, security isnt always the very 1st thing coded, maybe its the very next thing, maybe it comes 10 changes later.
Post reply on HN