back to article AI agents? Yes, let's automate all sorts of things that don't actually need it

The "agentic era," as Nvidia's Jim Fan and others have referred to the current evolutionary state of generative artificial intelligence (AI), is going to be a huge disappointment. The simple reason is that most people do not need AI agents to book their trips, order groceries online, or book restaurant reservations. Those who …

  1. Anonymous Coward
    Anonymous Coward

    There is nothing new under the sun...

    Automation has been pushed for at least two decades now, and this is just the same junk in a new hat. It's great for tasks that are a good for for automation, but ramming it in everywhere is a recipe for disaster.

    After all, at the end of the day a computer is simply a device for making mistakes faster.

    (And I've been automating tasks for 40-odd years, long before it was cool...)

    1. cyberdemon Silver badge
      Holmes

      Re: There is nothing new under the sun...

      > After all, at the end of the day a computer is simply a device for making mistakes faster.

      And AI is a device for making even bigger mistakes, ones that can't easily be found, corrected or exactly reproduced (unlike, say, a wonky reference in a spreadsheet) even faster

      It's not just that few people "need" AI, but few trust it, for good reason. Just because it "worked" today doesn't mean it will tomorrow, and when it doesn't, you'll have a hard time fixing it

      1. teebie

        Re: There is nothing new under the sun...

        Way too many people trust it. And it lets those people Dunning-Kruger faster.

    2. Anonymous Coward
      Anonymous Coward

      Re: There is nothing new under the sun...

      Hear hear. There are tasks that can (and should) be automated, and there are ones that require a human for decision-making and notice-the-odd-thing capability. Do NOT confuse the two, and do NOT attempt to automate a human-specific task.

      Coming from someone whose job title (Automation Engineer) actually contains the word automation!

    3. DS999 Silver badge

      Not sure I agree

      I would consider an "AI agent" as being effectively like a personal secretary or personal assistant, and that's different than "automation". Automation is something that doesn't require any intelligence at all. If I notice my milk is almost out if I have an Alexa or whatever I could say "hey Alexa add milk to my grocery list". Or maybe to add it to my next grocery order if I have groceries delivered.

      But automation falls apart if there is much required beyond that. Maybe I want to buy a new USB-C charging cube. There are a range of sizes, capabilities and prices so I can't say "order me a USB-C charging cube from Amazon" because its just going to pick one at random. If I look myself I would do things like restrict it by Prime shipping (if I need it quickly) 4 star ratings or above (for whatever their ratings are worth, I certainly don't want one that can't cross that low bar) and I'll have some minimum wattage and maximum price in mind so I'll go through them. Maybe I'll read reviews if I determine I need to know more about this and decide if I care whether it is GaN or not, see if there are any "big name" brands I can get rather than the alphabet soup of fly by night Chinese companies.

      "Automation" cannot accomplish that process for me. I would call something an "AI agent" when it IS able to accomplish that for me. Maybe it brings me a few alternatives showing a few representation options at different points along the spectrum, maybe it asks me a few follow up questions, etc. Basically the same as what a personal assistant would do if I was some celeb and could afford to pay someone $100K a year just to be at my beck and call for mundane tasks. The most important distinction is that "automation" doesn't learn. It just does what it is told. I would expect an "AI agent" worthy of the name to learn from working with me. That there are some tasks I assign it that I care about more than others, so some it would know just to do the best it can and not bother me with followups, others where I really want it get it right and to ask me as many questions as needed to get the optimal outcome.

      1. munnoch Silver badge

        Re: Not sure I agree

        All of that is true, and an even greater obstacle is the "unknown unknowns" when you start into a task. You can't give the assistant directions on an aspect of the search that you don't even appreciate yet is relevant. And by the time you've done enough research to figure that out then you've pretty much done all the leg work yourself so what's left for the agent to add?

        The other thing is that someone has to build all the hooks that these things cling on to. You're not going to book a trip without all the interfaces that booking.com already built out to the hotels and airlines and all the myriad search options and filters that they expose. An agent is an infinitesimally thin veneer on top of a shit ton of work that other people have already done.

      2. Truth1492

        Slightly narrow sided thinking.

        Most people don't understand anything about ai and how to get proper responses. Even with an AI so-called agent it still needs more than just go find me a USB c drive. The same way that if you use any llm properly you should start with the proper prompt format or what you think the proper prompt format would be. Asking a generic question from an entity or AI that has been trained on all knowledge or anything available on the internet is pretty f****** retarded.

        Without fine-tuning AI needs to be prompted to be fine-tuned in the short-term that can then be trained to learn long-term. I've developed several systems to do such things as this as well as cut the cost and compute down about 99%. Hopefully Elon musk takes me up on the offer I sent him the help finish off the rest of his competitors with a new complete training model.

        Also agents are the near future's future of what we will be needing in the future. Even now people are very narrow-minded when it comes to agents. They're thinking about ordering pizza when you can have an agent automate and give you 14 different views on anything that you would feed into it. the 14 was a generic number it could be any number of views that you prompt texted to learn before.

        I hope you have a fantastic evening

        1. IGotOut Silver badge

          Re: Slightly narrow sided thinking.

          @Truth1492

          So what you are saying is I have to very carefully state what I want, otherwise I may not get what I ask for?

          So if I want a a USB-C I need to go...I don't know

          Please purchase a 3m usb cable with a female type c connector at each end, with a black and white braided cord, that has at least a 4 star review rating with a minimum of 1000 reviews, each containing at least 6 words. Then please get the vendor to send it via the cheapest method, but to my alternative drop off point rather than my house. If this isn't in stock then please can I have either a 5m cable......

          Or I just go on a fucking website and go, click, click,click, done!

      3. O'Reg Inalsin

        Re: Not sure I agree

        for whatever their ratings are worth, I certainly don't want one that can't cross that low bar

        Automated AI bots burning the midnight to create plausible sounding 1 star reviews will soon destroy the plausibility of that assumption.

        May the strongest bots win!

    4. John Smith 19 Gold badge
      Unhappy

      Re: There is nothing new under the sun...

      True.

      This sort of stuff was in AI research since the early 1980's.

      EG Marcus's Wait-and-see-parser was the core of a diary management application split across a whole AI class.

      Likewise "Gus" out of Xerox PARC IIRC.

      when you think that about 100 words make up 50 percent of the English language internet (and a lot of the rest are in the 3 categories of Verbs, Nouns and Adverbs, with the Nouns splittable into whole sub-classes with sub-classable properties Eg Cities, rivers, people etc) yet it takes 30+ billion activations to process a single token in DeepSeek (according to their paper) you do wonder why a bit more focusseed effort couldn't yield quite large results.

  2. Anonymous Coward
    Anonymous Coward

    Agentic, AI and Arseholes all have one thing in common

    They all begin with A.

    As does avoid.

  3. Doctor Syntax Silver badge

    "...linked to Booking.com."

    The use of booking.com and the like exactly the sort of thing for which automation would be useful - to open the mailbox address dedicated to it long enough for any transaction-related mail through and then slam it shut again. It's a chore but an essential one. Spam filtering will just waste excessive amounts of CPU power and some would still get through.

    1. b0llchit Silver badge
      FAIL

      But it uses State-Of-The-Art technology that we all like and need. No reason to be negative. You know, you need it, you want it, you require it. It is shiny and new. It is the must have status symbol for any and all MBA to be proud of to show off to the chumps over a glass of some very expensive expense account draining totally worth it booze.

    2. UnknownUnknown

      Just getting the Spam filtering right in the first place Shirley…..

      It’s like flipping a 3 sided coin.

    3. weladenwow

      #!/bin/bash

      thunderbird&&echo

      pidVal=$(pidof thunderbird)

      sleep 600

      # 10 mins

      kill -s15 $pidVal

      #pipPIP!

  4. Like a badger

    Electric scooters have other uses, you know

    Like getting perps between petty crime scenes, keeping firemen busy, knocking over pedestrians, and tripping up those with sight problems.

    1. Bebu sa Ware
      Facepalm

      Re: Electric scooters have other uses, you know

      Like getting perps between petty crime scenes, keeping firemen busy, knocking over pedestrians, and tripping up those with sight problems.

      Totally concur.

      The typical rider apart from being arrogant and rude, usually has two perfectly usable components of an all terrain vehicle attached to their bum viz legs which they used more often their aforementioned bum would likely be less massive.

      Shank's pony got H.sapiens from the the middle of Africa to Tierra del Fuego in a surprisingly few generations so one would have thought Shank's would be more than adequate to pop down to the cornery to buy a litre of milk.

      When this blight was introduced into Melbourne some of the more enlightened resident turfed the mechanisms into the Yarra which apart from polluting a river wasn't such a bad idea. Being Melbourne I imagine their streets being littered with these gaudy and ugly objects offended their aesthetic sensibilities rather than any immediately rational motive.

      Just today an electric scooter Einstein decided to cool his overheating battery in his freezer and burnt out his apartment† block. Unit Fire: eScooter battery in freezer

      † unit

      1. collinsl Silver badge

        Re: Electric scooters have other uses, you know

        Here in the UK, specifically in London and some other cities, our pavements (sidewalks for left-pondians) are now blocked with "Lime bikes" - they're app-based hire bikes which can be left in designated "zones", which generally means people just dumping them in front of building entrances, in large clusters on convenient corners, and sometimes edging out into the road, making it even harder to drive past.

        And of course like in Aus there are people who throw them into rivers, onto canal towpaths, into people's gardens, onto bus stops, vandalise them, and generally abuse them.

        Boris Bikes (the Transport for London cycle hire scheme introduced under Boris Johnson's term as Mayor of London) are a better idea because you have to return them to a designated bike stand which clamps onto the bike and anchors it to the street - you don't finish paying until the bike is in a stand, so you're incentivised to leave it in the correct place (Lime Bikes should do the same thing but often their "zones" are nebulous), and it's much harder to do nefarious things with parked bikes or leave them in the way. The downside is that you have to put the infrastructure in place to enable a bike stand which reduces flexibility, but the parking "zones" have a similar effect, even though they can be created, moved, or destroyed at the push of a button.

        1. Like a badger

          Re: Electric scooters have other uses, you know

          All credit to those fine members of society who put e-scooters beyond further use. If this trend can be encouraged the commercial model for renting them will be destroyed and hopefully they'll vanish.

          Short of an outright ban, not sure what we can do about privately owned ones as they're already illegal to use in public in the UK. That seems little deterrent, nor are the regular fires and deaths around 1-2 a month in the UK at the moment, with ten times as many injuries. If anybody hasn't seen the violence and speed of an e-scooter or e-bike battery blowing up, then there's several scary videos, for example this one:

          https://www.bbc.co.uk/news/articles/c623938dq6no

          The unfortunate thing is that e-bikes and e-scooters are seen by the low carbon fascists as important tools for their glorious net zero future and are actively promoted by national and local government policies. The fact that these devices don't generally replace any car, bus or train journeys, and thus are a net increaser of energy use is beyond the eco zealots.

          1. whiteknight

            Re: Electric scooters have other uses, you know

            "low carbon fascists"????

            That would be the scientifically literate community who realise the planet is on the cusp of total collapse.

            Unlike the denialist brain dead illiterates.

  5. Headley_Grange Silver badge

    It's a bad enough idea to start with but, as the author says, it's little more than a hands off google search, click through and autofill as described. Imagine what that'll be like once it's owners squeeze every penny out of it and services can pay to be the top AI selection. Let AI book that flight for you and get a 3-leg hop on three Ryanair flights to an airport only just in the same time-zone instead of a direct flight on an airline that's paid less to be featured and then find out there's no refunds.

    1. Anonymous Coward
      Anonymous Coward

      Nothing that can't be automated.

      Window 12: "You had a J Arthur three days ago, you probably deserve another? Try these images that our AI-curator feature has downloaded to your machine, based on Recall's tracking of your previous searches"

      1. m4r35n357 Silver badge

        Are "hallucinations" a good thing for wank recommendations?

        1. collinsl Silver badge

          It'll probably end up giving you images of Marvin the Martian roasting a cow over a fire thanks to a hilarious misunderstanding.

          1. Strahd Ivarius Silver badge
            Mushroom

            Where's the kaboom? There was supposed to be an Earth-shattering kaboom!

    2. GoneFission
      Devil

      And the "assistant" has the great side effect that it can embed itself into every aspect of your digital and physical life to pipe information back to the collection entity of choice. We noticed you had 15 minutes of downtime based on location and activity data. Your employer has been notified and the time logged for review, since you could have used this opportunity to answer AI-generated emails in your inbox instead. Sign up for our CorpAI Assistant Pro Plus feature today to allow offloading of some of the more time-consuming tasks to our AI Agents!

      While you continue waiting for that 3-leg flight, why not enjoy & rate some Sponsored Podcast Content by our Premium Partners to bump up your CorpAI Smart Score? "Effective Time Management At Work Brought To You By Billionaire Visionaries" has been inserted into your playback queue based on recent behavior. Please give your full visual and auditory attention to the twelve Premium Partner advertisements that precede the selected content. Deviation in attention or a lack of "Like & Subscribe" activity will result in an automatic subscription to our ad-limited Ultra Premium Tier.

      Your plane has landed and is currently preparing for boarding. Would you like to automatically upgrade to Expedited Pro Boarding Plus for priority seat assignment? Say "no thank you but I really appreciate it" to decline. Sorry, I didn't quite catch that, can you repeat that? Our servers are currently unavailable. The Expedited Pro Boarding Plus charge has been applied. Sorry, your selected flight is not eligible for this service at this time.

  6. Pascal Monett Silver badge

    "We know bad actors may try to misuse this technology."

    But we're ready to make money off of that as well.

    1. Yet Another Anonymous coward Silver badge

      Re: "We know bad actors may try to misuse this technology."

      I thought AI was going to replace bad actors with CGI ?

      1. Headley_Grange Silver badge

        Re: "We know bad actors may try to misuse this technology."

        They don't need to replace the bad actors - CGI costs more than the wages of a bad actor. It's the good actors they want to replace.

      2. Androgynous Cow Herd

        Re: "We know bad actors may try to misuse this technology."

        It’s actually a warning to keep an eye on Vin Diesel.

        1. LBJsPNS Bronze badge

          Re: "We know bad actors may try to misuse this technology."

          Of course. Who wants Diesel fueled AI?

  7. phil_4

    Everything I've seen of Agents so far is just putting a pretty end on some sort of menu or interaction. The example was to help with helpdesk tickets. The AI was able to work out what the user was asking for, and could then... well that's the point. It couldn't do anything out of the box. Some poor sap had to wire up all the "solutions". Then the AI just ran whatever solution it was given. In some ways this is great as it automates away a lot of repetitive helpdesk tickets. However 1) If you had some repetitive tasks anyway, you should have found another way of removing the need for a helpdesk ticket before hand 2) The AI really isn't doing any more than providing a nice way of asking for things.

    Same with RAG training stuff, searching your own docs and data. It's all just a more natural, easier to use front end on things. It's not actually doing anything clever entirely of its own accord.

  8. Mike 137 Silver badge

    A certain kind of business...

    "... it's less clear why businesses might want to become endpoints in OpenAI's ecosystem and surrender direct contact with customers"

    Unless they're the increasingly common kind of business that thinks customers are a nuisance except (briefly) when their wallets open. Of course you wouldn't want to do business with them, but once there's an "AI" intermediary you might not even realise you're doing so -- a further extension of the Amazon marketplace, where even now you don't really know whether the notional business actually exists.

    1. doublelayer Silver badge

      Re: A certain kind of business...

      Another type is companies that should have created a better UI, didn't, and hope that the LLM will do it for them. For example, some sites that are mostly used as price comparison sites but do their best to hide it. They know that the users are coming to see a list of prices and, where the products differ, the relevant details, but they won't make as much money giving them that as giving them a list of suggestions ordered by how much money the site operators can get for the user choosing them. Those sites can still be used to get the data the user wants, but the UI makes it trickier. And although my example posits such a site where the choice to do it badly was deliberate, that's not all of them either. Some of them are trying to provide the best UI, but the things they're comparing are complex enough that you can't just give someone a list of choices by price; they have to do some comparison themselves with the data you've given them. A good UI would give them filters that they can use to narrow down their choices, but that would take manual effort to design those filters. Theoretically, a functioning LLM would do this for them. In practice, it will mostly break or skip important things, but that's for users to notice rather than the site's maintainers.

  9. find users who cut cat tail

    Human brain

    A CPU can calculate sin(0.176354405021) to fifteen decimal places in nanoseconds. The energy necessary may be a few µJ.

    It would take me hours and I would probably make mistakes. The energy consumption is at least 10 orders of magnitude larger.

    Human brains can do lots of things efficiently, but performing mathematical computations is exactly the thing which they cannot do. The cited article confuses ‘doing things’ with ‘performing mathematical computations’. It is like saying electrons can do quantum mechanical computations very fast – when they are just being electrons.

    Now when you ask ChatGPT to calculate the sine, that's a different matter.

    1. Bebu sa Ware
      Coat

      Re: Human brain

      "Now when you ask ChatGPT to calculate the sine, that's a different matter."

      Which is the gist.

      Humans on the whole couldn't calculate trigometric functions to fifteen places but clearly they have worked how to. Various bits of numerical analysis have reduced these tasks to fairly simple steps that could be performed with a four functional calculator as I believe mathematical tables were produced by teams of human "computers" using mechanical calculators.

      Most trig functions' series expansions converge fairly quickly so anyone with a calculator that could perform the four functions to say 20 significant figures could probably knock out 15 places of a sin() in reasonable time.

      ChatGPT doesn't know how to calculate a sine because it doesn't know anything but I am surprised that the pattern isn't recognised and off loaded to an online calculator or Mathematica.

  10. Neil Barnes Silver badge
    Mushroom

    AI simply isn't welcome in human spaces

    Or indeed, except by the idiots pushing it, anywhere else.

  11. OldSod

    (some) hidden dangers of agentic/generative AI

    We get answers, but we don’t know how good the answers are:

    When I do my own research/thinking, I gain insight into the structure of the problem and its context along with the answer space; if an AI tells me an answer I have much less insight into how good that answer is compared to other possible answers.

    We have answers, but we don’t know anything:

    There are lots of ways to evaluate choices and make tradeoffs. Some analytic disciplines such as systems engineering spend a lot of time developing systematic qualitative and quantitative approaches to analyzing choices and evaluating tradeoffs. As the analysis/evaluation tools are applied to a specific problem, the outcomes may not be determined completely by the systematic tools, but by the insight gained by the application of the tools. When someone/something tells us the answers, even if they can explain the rationale behind the answers, it is that entity's rationale, and not our own that we have developed by our own analysis of the problem.

    We have answers, but we don’t even know what the questions are/were:

    An agentic AI that is tapped into all of our context and interactions may make recommendations to us, telling us to do this or that, but we may not even know what motivated the recommendations.

    We get our answers from a middleware that is tapped into every aspect of our life and thinking but ultimately reports to someone else:

    At one point it was considered alarming that a movie rental business might come to know something about its customers by virtue of their access to a list of all of the movies the customer had rented… how quaint this seems in light of the near constant data collection that is taking place as we concentrate knowledge of everything we do in the hands of credit card issuers and Internet behavioral surveillance systems like those operated by Google and Facebook. Now we risk having an even greater awareness of all that we do in the hands of the operators of the AI agents. Jarvis was theoretically intensely loyal to Tony Stark. Agents as a service will not be loyal to the users.

  12. glennsills@gmail.com

    No one wants to pay for AI

    OpenAI needs as many stories as possible to keep that investor money coming in. I suspect, that OpenAI, like Microsoft, Google, Apple, and Meta are finding it hard to sell AI products to end users. Giving stuff away until AI catches on is not an option for OpenAI. For that reason, they keep coming up with new applications that people *might* want.

    1. abend0c4 Silver badge

      Re: No one wants to pay for AI

      Part of the problem is that there are so many intermediaries there's no money left - and the final product is unknown to any of them.

      Booking a rental car? The website hosting the search service will take a cut, the 'white label' search service provider will take a cut and direct you to an aggregator who will take a cut and pass you on to a branded franchise operator who will take a cut and you finally get to the car hire company who'll charge you a fortune in insurance and extras to recoup the lost revenue from the booking - which the 'intelligent' agent will take no account of.

      The reality is that technology isn't putting us in touch with actual businesses, it's just creating an ever-lengthening line of outstretched hands.

  13. nautica Silver badge
    Meh

    From the last sentence of the article...

    "Someone is sure to appreciate it."

    But just not me. ["But just not I", for all you grammar-nazis]

  14. Anonymous Coward
    Anonymous Coward

    Let's stay grumpy old SysAdmins

    First we ignored it

    Now we are laughing at it

    Then we will fight it

    The we will submit and start using it

  15. Wiretrip

    Anyone who has any doubt about how frighteningly stupid 'Agentic AI' is should watch this fantastic talk by the great James Mickens:

    https://www.youtube.com/watch?v=ajGX7odA87k

    1. raoool

      Ain't nobody got time for dysentery

  16. tiggity Silver badge

    ..Unfortunately

    ..Unfortunately "AI" does not need to be much good to gain commercial traction.

    Plenty of companies happy to have mediocre ""AI" help bots / automation" and less human customer support staff if they can save some cash .. the "bots" do not have to be as good as the humans, so long as the bots can successfully process a cost effective amount of tasks / customer queries*, leaving the humans to deal with more taxing tasks then its a win, and the more staff can be replaced the bigger the win.

    * potential for "bots" to be faster, so another reason why don't need to be that good to be a commercial win as can essentially act as a speedy triage mechanism, dealing with the low hanging fruit (easy stuff) rapidly & passing the more difficult work up the chain (e.g. If they can deal with commercial equivalent of L1 and some L2 support work then that company gets a decent bottom line boost).

  17. O'Reg Inalsin

    AI is already helping busy doctors automate the process of challenging insurance rejections written by AI. This kind of perpetual motion system can only be made even more efficient with AI agents. Who would have dreamed of perpetual motion in our lifetimes? What a great time to be alive (*exempting those with serious illness)!

  18. teebie

    "Electric scooters are useful for certain scenarios [...] Operator is like that"

    It will be seemingly only used by over-confident arsehats, and a scourge for everyone else

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Other stories you might like