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

What, exactly, is an 'AI Agent'? Here's a litmus test

tines.com

Post reply on HN
Print view View original

31–40 of 45 posts

Re: What, exactly, is an 'AI Agent'? Here's a litmus test

#31

Post by swyx »

falcor84 wrote: post #19
janalsncm wrote: post #13

It’s a bit amusing that so much ink has been spilled over what the definition of an “AI agent” is. I don’t care. I care what your software can do. I don’t care if it’s called AI or machine learning or black magic. I care if it can accomplish a task reliably so that I don’t have to do it myself or pay someone to do it. We had the same argument about 3 years ago when everyone started calling things “AI”. They use LLMs…

It's fine that you as a user of these systems don't care, but nevertheless this is useful terminology for people looking to design such systems.

agree. a really good definition leads to a really good mental model which leads to really good design. however people can get in a penis measuring contest over definitions too which is often not great

Re: What, exactly, is an 'AI Agent'? Here's a litmus test

#32

Post by simonw »

This Hacker News thread is already a great example of what always happens when this topic comes up: I count SIX definitions of agents in this thread already, each slightly different from each other but each expressed with a confidence that suggests that the author thinks their definition is clearly the correct one.

(The OP did that as well.)

Re: What, exactly, is an 'AI Agent'? Here's a litmus test

#33

Post by TeMPOraL »

> Does the AI system perform actions under its own identity? If it does, it’s an agent, and the audit logs will name the agent itself. And if it doesn’t – like most copilots or in-product assistants – it’s not.

God please no, let's not normalize this idea.

1. That's not really a good definition of an agent;

2. The only agents I care about are agents acting under my identity, for me, in my interest. You know, like browsers were supposed to - that's where the name for "User-Agent" header comes from. But in short - whether I'm accessing your service directly or using an agent (AI or otherwise) to do it for me, is none of your business. Letting service providers differentiate based on that was a cardinal mistake of the early Web.

Re: What, exactly, is an 'AI Agent'? Here's a litmus test

#34

Post by aaron695 »

bhouston wrote: post #2

"Does the AI system perform actions under its own identity?" I don't agree with this definition. I view an agent has having the ability to affect the world, and then sense how it affected the world and then choose to make additional actions. Thus there is an act, sense, re-act feedback loop going on that does not require a human to mediate this. This to me is an agent. "But why isn't, say, ChatGPT an agent?" ChatGPT…

[dead]

Re: What, exactly, is an 'AI Agent'? Here's a litmus test

#35

Post by 1as »

simonw wrote: post #32

This Hacker News thread is already a great example of what always happens when this topic comes up: I count SIX definitions of agents in this thread already, each slightly different from each other but each expressed with a confidence that suggests that the author thinks their definition is clearly the correct one. (The OP did that as well.)

Hey, OP here. I’ve really enjoyed your work on AI agent definitions. So thank you for reading!

I don’t actually claim that ours is necessarily the correct answer for everyone – it’s our own. But I believe it is at least _an_ objective definition. Other definitions I’ve seen have been murky and subject to interpretation.

Re: What, exactly, is an 'AI Agent'? Here's a litmus test

#36

Post by janalsncm »

pixl97 wrote: post #26
janalsncm wrote: post #13

It’s a bit amusing that so much ink has been spilled over what the definition of an “AI agent” is. I don’t care. I care what your software can do. I don’t care if it’s called AI or machine learning or black magic. I care if it can accomplish a task reliably so that I don’t have to do it myself or pay someone to do it. We had the same argument about 3 years ago when everyone started calling things “AI”. They use LLMs…

>I don’t care. The particular problem with poorly defined definitions is they cause a lot of spilled ink later on. For example the term AGI. Or, even deeper, the definition of intelligence, gets debated again and again with all the goal post dragging one expects these days. Even breaking out simple categories can help like Type I agent: Script driven, uses LLM for intelligent actions. Type II agent: LLM driven, uses…

Now we’re talking. That’s a useful framework because it acknowledges there are gradations of independence. It’s not an all or nothing thing.

Re: What, exactly, is an 'AI Agent'? Here's a litmus test

#37

Post by bhouston »

simonw wrote: post #30
bhouston wrote: post #2

"Does the AI system perform actions under its own identity?" I don't agree with this definition. I view an agent has having the ability to affect the world, and then sense how it affected the world and then choose to make additional actions. Thus there is an act, sense, re-act feedback loop going on that does not require a human to mediate this. This to me is an agent. "But why isn't, say, ChatGPT an agent?" ChatGPT…

Is ChatGPT with its Code Interpreter tool an agent?

Good point. It is a bit of a grey area. It is acting, then executing the code, and sensing the results, and then making changes. So in that sense it is an agent, but it is a little self-contained.

In a way thinking is sort of agentic in a way, it is talking to itself and sensing it and deciding what to think next...

Re: What, exactly, is an 'AI Agent'? Here's a litmus test

#38

Post by antonkar »

The Big Bang is maximally agentic but has zero intelligence (it has maximal future potential intelligence though). Current definitions of agency are way too narrow and unphysical, so:

Agency is time-like, energy-like, choosing, changing shapes or geometry of the world and the agent itself. It’s GPU computations. Explosions have a lot of agency.

Intelligence is space-like, matter-like, the static geometric shape like an LLM (it’s basically a bunch of vectors). It’s a file. The static final 4D spacetime of our universe has a lot of intelligence but zero agency, because it’s static.

Maximal intelligence+agency is the static spacetime of multiverse (=max intelligence) which can change its shape in an instant (=max agency, shape-changing ability).

Same way we have e=mc2, we have agency = intelligence * constant.

Re: What, exactly, is an 'AI Agent'? Here's a litmus test

#40

Post by Terr_ »

float4 wrote: post #20
andy99 wrote: post #7

Anthropic has a definition: Workflows are systems where LLMs and tools are orchestrated through predefined code paths. Agents, on the other hand, are systems where LLMs dynamically direct their own processes and tool usage, maintaining control over how they accomplish tasks https://www.anthropic.com/engineering/building-effective-age... While I know it's a marketing term, I think a good distinction is that agents hav…

I should have read this 12h ago! This afternoon, I tried to create my first simple agent using LangChain. My aim was to repeatedly run a specific python analysis function and perform a binary search to find the optimal result, then compile the results into a markdown report and export it as a PDF. However, I now realize that most of these steps don't require AI at all, let alone agents. I wrote the full algorithm (in…

That reminds me of another recent submission that seems relevant:

"Don’t let an LLM make decisions or execute business logic"

319 points, 168 comments, 1 day ago - https://news.ycombinator.com/item?id=43542259

Post reply on HN