'No One Knows What the Hell an AI Agent Is' (techcrunch.com) 40
Major technology companies are heavily promoting AI agents as transformative tools for work, but industry insiders say no one can agree on what these systems actually are, according to TechCrunch. OpenAI CEO Sam Altman said agents will "join the workforce" this year, while Microsoft CEO Satya Nadella predicted they will replace certain knowledge work. Salesforce CEO Marc Benioff declared his company's goal to become "the number one provider of digital labor in the world."
The definition problem has worsened recently. OpenAI published a blog post defining agents as "automated systems that can independently accomplish tasks," but its developer documentation described them as "LLMs equipped with instructions and tools." Microsoft distinguishes between agents and AI assistants, while Salesforce lists six different categories of agents. "I think that our industry overuses the term 'agent' to the point where it is almost nonsensical," Ryan Salva, senior director of product at Google, told TechCrunch. Andrew Ng, founder of DeepLearning.ai, blamed marketing: "The concepts of AI 'agents' and 'agentic' workflows used to have a technical meaning, but about a year ago, marketers and a few big companies got a hold of them." Analysts say this ambiguity threatens to create misaligned expectations as companies build product lineups around agents.
The definition problem has worsened recently. OpenAI published a blog post defining agents as "automated systems that can independently accomplish tasks," but its developer documentation described them as "LLMs equipped with instructions and tools." Microsoft distinguishes between agents and AI assistants, while Salesforce lists six different categories of agents. "I think that our industry overuses the term 'agent' to the point where it is almost nonsensical," Ryan Salva, senior director of product at Google, told TechCrunch. Andrew Ng, founder of DeepLearning.ai, blamed marketing: "The concepts of AI 'agents' and 'agentic' workflows used to have a technical meaning, but about a year ago, marketers and a few big companies got a hold of them." Analysts say this ambiguity threatens to create misaligned expectations as companies build product lineups around agents.
yeah, but OpenAI still thinks (Score:5, Interesting)
I know the answer (Score:3)
It's something that your business must buy RIGHT NOW or you are behind the times. It costs exactly 1.5% of your gross annual receipts. Just think of the *future savings* this will bring you when you no longer have to employ people to do whatever this is. You will, however, have to hire expensive experts to verify the system is operating within legal parameters. We don't know why the AI insists we run the blood of infants through our datacenter cooling systems, but we'll adjust your bill for that if it ever becomes a problem.
Re: (Score:2)
Waaaah, my FOMO is soooooo bad now! (Or would be if I were an idiot of the type you implicitly describe...)
Re: (Score:2)
Have you met a CEO? They're replaceable idiots. One can be killed shortly before a major shareholder conference and the business just keeps on as if nothing happened.
Re: (Score:2)
Oh, yes. Actually a business may do _better_ after a CEO keels over dead. Too many CEOs think they are critical. That is true, a CEO is a critical failure factor. A CEO very rarely is a success factor and almost never a critical one.
One such meeting I had was as an external tech evaluator. The CEO did insist aggressively how great their product was and did not listen to anything. The CTO was more knowledgeable and actually gave me information, but he was completely under the spell of the CEO. My prediction
It's all a bubble anyway. (Score:2)
All those agents are focusing on trying to do "WOW-factor" things but it's nothing more than a marketing stunt to rather than doing something actually useful. Whatever keeps the bubble growing... To quote myself from a comment under a different Slashdot article:
"People don't want "AI" things that are useless. They want dumb, simple things that are useful, consistent and reliable."
Re: (Score:2)
Companies are mostly fine with this lie. Because they can prop themselves up with the same lie by saying that they're using said technology and their stock price will ride the bubble a bit too.
The summary portrays confusion, but... (Score:4, Interesting)
There's not really a conflict between the various "definitions" of AI agents. At their core, agents do things, while AI chatbots can merely talk (or type).
Types of agents? Fine, what's the problem? Some agents can tweak OS settings, some can open apps or navigate (GPS), some can auto-browse web sites, some can write code (directly in your IDE). These are all types of agents. The fact that there are different categories, isn't any more confusing than that there are different categories of software in general.
In other words, clickbait.
Re: (Score:2)
Indeed. Although the activities will be more likely break your system, steal your data and money and generally do stupid things instead of just saying them.
Re: The summary portrays confusion, but... (Score:2)
Re: (Score:2)
You are projecting. You should stop doing that, it is not healthy.
Re: The summary portrays confusion, but... (Score:2)
Re: (Score:2)
Maybe, it depends on who you do business with. As with any business you transact with, you should do your due diligence and not use products from businesses you don't trust enough to be safe from things like stealing. AI has nothing to do with it. A malicious chatbot will steal your stuff just as quickly and easily as a malicious "agent."
Re: The summary portrays confusion, but... (Score:2)
Re: (Score:2)
It's not like there weren't tens of agentic open source frameworks to do things in-house.
is it because... (Score:2)
Re: (Score:2)
That one is easy: Artificial moron that uses statistics and gets mistaken for possessing reason by natural morons.
Ads (Score:2)
Re: (Score:2)
Congratulations, sir.
Look at the latest google television ads for the pixel phone. 100 years ago you would set a 5 second shutter timer on your camera and run in front of the camera to get into the group shot. Now supposedly you're too dumb to do that, you need "AI", so that you can take the shot, then... what I don't know (or care) how it works... you take the group shot, then add yourself into the picture after?
It's clear the target market for the Gpho
A much abused term (Score:2)
I was just in a meeting on a research project, where every diddly little program was an "agent". Sounds cool? Helps get funding?
Ah, marketing. Selling what doesn't exist. (Score:2)
How many folks here have been in the situation of trying to figure out what the hell marketing just sold, so they can build something that matches?
Re: (Score:2)
Me? Never. And if it ever happens, I will instead produce a nice, well-written (no AI usage) letter of resignation. Although with my current jobs, something as bizarre as that is very unlikely to happen.
My sympathies to anybody faced with such a crappy situation when they cannot realistically resign their position.
Re: (Score:2)
yes. +1. Not just once either. One time the new chief marketing person, who had just left McDonalds to help us "find our happy meal" decided to market a feature we implemented, but described it working completely different from what it did. We had to figure out if we could actually use the data we were pulling in to do that technically and legally (flight tracking radar information) within 5 days. We ended up making it work, but it was a long 5 days.
What? Isn't it clear? (Score:2)
An "AI Agent" is an agent sent to spy on you, manipulate you and and steal your money by way or purchases you would otherwise not have made.
Think "agent" in the sense of somebody working for Putin in the US government and you get the idea.
What the hell... (Score:2)
'No One Knows What the Hell an AI Agent Is'
Nice, that puts them in the same category of undefined crap as CRT, Woke, and DEI. No two Trumpsters seem to be able to agree on what those are either.
pretty simple ... (Score:2)
A software coding agent should do the job of a system software engineer. It'd be a co-worker not a tool. If it's good and fast enough you can fire all the human workers.
For example: the project at work precreates containers at a constant rate. Precreated containers depend on the state of the system, so they age out after awhile as the system changes. If the rate were dynamic based on past behavior, it'd use less resources most of the time and avoid running out during emergencies. There's also some control c
Is it just the usual AI but using JSTOR or somethi (Score:2)
Let me try to answer (Score:1)
It's an irritant that provides minimal value at maximum hardware and energy cost.
It's a strange situation (Score:2)
Genuinely useful work is being done, but it's still preliminary and not yet ready for widespread deployment
Meanwhile, the hypemongers and marketoids are inventing words and preparing to sell expensive products that will probably suck
I'm optimistic that useful tools will be developed, but we will have to swim across a river of crap to get there
Not a problem, just growing nomenclature. (Score:1)
This is just old news with a new skin. What is a program, platform or a system? Are you building microservice or monolithic architecture? How does serverless skills like Lamda/Fuctions fit into a program workflow?
If history is any lesson, good luck applying today's terms to tomorrow's architecture.
I work in this area for a big 3 cloud. We are trying to understand how to best describe our platform design. Is each step/skill a modular agent, or do you wrap everything into an agent that represents an enti
Re: (Score:2)
Yes, and like all specializations, AI is rapidly developing its own vocabulary both the allow experts to communicate with one another AND exclude laymen from understanding what's being said. To the outside world, such specialized vocabulary is useful because it entitles insiders to take more money.
I know what it isn't (Score:2)
Could I have been using one? (Score:2)
those from the '90s remember... (Score:2)
the late 90s was a period where 'agents" were all the rage in popular tech media and in the lexicon of large-scale software architectures.
in the military this was reflected in the idea of various elements able to handle telemitry be able to exchange information securely (e.g., one helicopter on one side of a mountain range could echo what its radar is seeing to another chopper on the other side of that range. the idea of 'agents' was to exchange that information without interactive efforts by either pilot.
i
No one CARES what the hell an ai agent is (Score:2)
Well defined (Score:2)