Maybe I get it wrong. I feel agents are too "fluid". To unstable, to unpredictable. There is still so much stuff that requires simple, predictable code, small db, simple front-end. I think some people want to apply heavy, resource consuming AI everywhere, like once it was with blockchain technology, which just doesn't make sense.
but DeepSeek R1 shows that models can be tuned to consume (and then require) less resources. Just because the typical American businesspeople are gluttonous asshats, doesn't mean they're going to dominate the market. Eventually they're going to turn their eye back to cutting costs (instead of investing) and that's when they'll care about efficiency again. Right now basically everyone is bullish on spending money on AI. So efficient models are not their highest priority. Capability is.
"Simple predictable code" is called "tools". This is not going away. I think as AI is optimizing "itself" it will write more tools too. The problem defines the solution! No one hunts ducks with howitzers!
@@jsbgmc6613 In the agentic architecture, agent is the one that makes the desicions according to some process. An LLM based agent will always be unpredictable (even if it has these tools) so those process are in the end better executed by deterministic software and databases.
If you don’t understand your current workflow then how you gonna come to a conclusion things are too fluid. Nothing we or the AI are building is meant to be perfect it’s meant to be adjustable as per needs arise.
Human: Hey Agent, show me a report of today’s sales. Agent: Sure human, here you go. Human: It’s missing totals. Agent: Sorry about that. Here you go. Human: These numbers don’t look right. Are you sure about them? Agent: You are right. Sorry about that. Here you go. Human: Teach me SQL.
Spot on ... People seem to easily forget it's just a word prediction engine 😉 There's gonna be more crap out there and senior experts still have value.
Yeah.. what they talk about sounds like artificially sustaining scarcity, by making the agents proprietary. Rewarding contribution and upholding attribution is one thing, locking down the software and using it as currency is another.
Why do you continue to act like anything these CEOs say is some profound prophecy for the future? You do realize that they’re MASSIVELY financially incentivized to make these absurd claims right? More hype, equals higher stock prices, regardless of whether or not what they’re saying is remotely true.
??? Agents are still software. Love the channel Matthew but some of your titles are clickbait. Everything is not STUNNING in caps, shocking, yada yada. One can talk about incremental AI advancements without resorting to sensationalizing it other than moments it really is something spectacular (e.g. Stargate Project). You'll get new users with clickbait titles but the more you do it the more you'll annoy your regulars. I already dumped a few other AI news channels for clickbait practices. Content creators who respect their viewer's time focus on accurate information not misleading tricks to get hits. Please don't go down that road because I really like this channel
I'm with you. I have unfollowed so many channels with great potential because of the clickbait practices. It looks dumb and made for dumb people. This one is testing me lol.
@@danskyder1564 "bycloud" and "Matt Wolfe" are really nice ones. Also "AI Search" (it is kind of clickbaity sometimes, but still pretty good in its content).
The problem with hiring a person with a "basket of agents" is that this is quickly going to go the way of Prompt Engineering as a career. The AI simply performs at a higher level than humans, and at a fraction of the time. Here is a quick example. I chose n8n as that is what I happen to have running at the moment, but it could be any framework/orchestration layer. Prompt: "I need to create an LLM agent that can process Stripe Payments. I would like to use n8n as the framework. " Response: "Creating LLM Agent with n8n and Stripe Thought for 2m 46s" Initial Output: "Below is a conceptual approach to building an LLM-driven agent that can handle Stripe payments using n8n..."
@CrispinCourtenay In addition, LLMs are advancing so fast, an agent built on GPT-4 won’t work on O3. Agents quickly become outdated and will require huge amounts of work to keep up to date, and testing them is going to be a nightmare.
Exactly. It's a joke, which is why Satya is laughing. The only people who will be making bank in an "agentic" future are the companies powering those agents. Nice ad btw.
@ Sure you can build agent, upon agent, upon agent, to upgrade and test each revision, but every layer you add, the complexity balloons and the cost to get the expected result increases inference time. It still costs a lot of money to keep that going. Sure it can be done, but it isn’t necessarily easy or cheap. Depending on the problem being solved, the cost might be justified, but for a person with 100 agents in their toolbox might not be a good value.
You really think companies will just let you walk away from them with agents you created while working there lol. This is obviously best case scenario for the salary worker.
Your agents exist in your mind! The workflows you build through the expertise you acquire over time, along with the ability to build new ones, are what truly matter. Adapting an agent to a new context is equally important. The only question is how the job market will remain open for juniors so they can gain enough experience to create agents. But keep in mind that this remains a short to mid-term perspective, valid only as long as AI has not yet achieved the ability to perceive the physical world and model workflows by itself.
Yeah. Companies have a cow now when the sales guy takes the customer contact list he built. The agent is IP of the business. Go build it again somewhere else.
I... I... I don't even... you do know the dude telling us the future of work will be wholly dependent on the number of created / owned agents in your resume is the same guy as the one who is selling the agents, right?
Not just that. How many agents you are bringing along.. I would imagine your previous company having serious issues with you re-using the intellectual property that you built for that company.
That doesn't mean he doesn't have a point Much like your resume will include examples of your coding it seems fairly obvious that building good agents will also be relevant
Like a lot of people here, I'm diving deep into Ai, including LLMs, tts, image and video generation. I'm embracing the technology as much as I can. BUT, I'm still deeply concerned how gleefully we're making human beings obsolete.
You probably should be. When the Klarna CEO replaced thousands of CS reps with AI, someone asked him in an interview (and I'm paraphrasing)- "Isn't that kind of a bad thing too?" His response was along the lines of "Only if you're a human"
The issue doesn't lie in our humanity; it lies in our economic system. It turns us into mere tools designed to generate profits for entrepreneurs and business owners. However, with the rise of AI, the playing field is leveling. Now, anyone can become an entrepreneur, and those with technical skills will have an edge in creating and delivering greater value compared to those without. In the end, if AI makes it easier to become a marketer, embrace marketing; if it simplifies coding, learn to code; if it streamlines content creation, dive into writing or video production; if it automates data analysis, explore data science; if it enhances design tools, master graphic or product design; if it improves customer interaction management, focus on customer success; and if it facilitates project planning, step into project management. AI is unlocking opportunities in every field-it's up to us to seize them.
@@TawfikAymannothing you said makes any sense if they plan to replace all those jobs with AI. I don’t think you understand what the Nadellas and Zuckerbergs are sayung
Great video. I was a process control engineer I would always stress test my projects just to make sure it would work under all conditions. But with my involvement in the digital marketing weekly income, has been life changing. Regardless of how bad it gets on the economy.
I’ve raised up to 460k trading with her l no longer pay rent, I own two apartments, a car and also investing in real estate. Myself and my family we are extremely happy🙏❤️
I believe that we still do not fully understand it, databases and programs exist today because WE HUMANS are not capable of remembering large volumes of information and displaying them in a conversation, but imagine a SUPER HUMAN, capable of receiving A LOT OF INFORMATION , ORGANIZE IT IN REAL TIME and SHOW YOU on a screen in real time, programs like WE KNOW THEM, are not required, because there will literally be SOMEONE, an AGENT capable of managing all that information and showing us it in real time. I'm going to compare it to something absurd, but in Bible times they talk about chariots of fire, how would someone who has never seen a spaceship flying something describe it??? How would we describe something that is capable of doing this if we have not seen it yet... that is why we continue talking about programs.
So, the next frontier, can you ask the right question? For the time being, programmer become tester, and then become product manager. Later on, you just think before services/products will come to you. All the middle tiers, what are they for? They are for to be replaced by AI/robots. You do, however, to understand how the database, logic, interfaces and everything work together conceptually. Your neuron should be busier, not your finger, not your foot, not your arm. Everything in between, is how do we get there faster. I didn't think I have a chance to see that day. Now I am not sure.
I'm writing a new CRM right now for a business. They have an old legacy database we need to incorporate. There is no short cut around having to build an interface for example for our sales people to enter in who they talked to today, what they did, and what projects they are interested in. And accounting needs hard numbers to be entered into our system. I cant just invent an agent that will manufacture all of this, with these fuzzy guidance properties, like 'lets keep track of customers we talked to today'. I think there will continue to be a need for the concrete foundations these things sit on, the hard core database design, the actual UI form used to enter hours, addresses, names, phone numbers. Once all that boring stuff is done and is nailed down, then the fluffy agents can come along and give you the generalizations of everything to help guide you. But its a long way from, 'create database' and 'new project' in Visual Studio, to asking AI, 'what were our sales today'.
If SaaS is dead, then there is only AI and you can pretty much sell the entire sector short: the economy will not survive this amount of value being destroyed from the money markets.
Destroyed?? nah, that layer of the technical pyramid is simply no longer the top layer. Agentic development is the next layer and builds on top of the old layer. Everyone who shifts gears and adds value to the agentic ecosystem now will profit, the stubborn will find their old skills quickly become diluted.
Actually he is saying business logic will be implemented by ai models instead of well defined structured backend code, its like classifying red and green balls using ai models instead of checking pixel value of the colour
Completely sceptical too. AI that is doing tasks inside systems is not coming soon, as actually it is too much error prone. Some company had already close their AI chatbot because of the lot of mistakes AI is doing. Imagine an AI doing thing automatically won't come this year with the use of LLM. It will stay as a very good tool to help users for the moment.
Yea... If my agents are that sought after... It would be because they were actually useful.... So.... If they are useful... Meaning they perform tasks that lead to revenue generation..... What in the actual fuck would I need a job for?
Why does anyone hire a plumber when they have plumbing issues? Why does anyone hire an accountant when their taxes are due? Why do people hire someone to assembly their flatpack furniture? etc. We all have the ability to go out, learn these skills, and perform the work ourselves. Yet there are industries which cater to these skills.
I know. Too much socially organised denial. I'm paid 5 days per week. 2 of those are done now with AI and the other 3 I spend learning about AI/agents etc. The current talk is that plumbers (for example) will earn more than software/IT folk. Thing is, there will be mass migration to the manual jobs that will be last to be replaced. Supply and demand, those workers will be on minimum wage before long. AI is a slow erosion of a workforce. Governments are not really acting fast enough to put a framework in place to pay a social allowance.
@@niv8880manual jobs will last longer, but not much, when we reach AGI it will be able to create robots thousands of times better than we do, so it could happen all within 1 to 2 years, in the first one office jobs will disappear, and in the second one the manual jobs will be gone
Nobody gets anything. Only the lucky one who owns an AI work force. They have whatever they want. Whatever they decide to give you is what you get. If that is nothing, then you get nothing.
@ I just wish everybody would understand that AI doesn't fire people; people fire people. Unfortunately, their greed makes them oblivious to the fact that by disrupting the humanity, they will also suffer the consequences. Because they are a part of this humanity. And humanity is a sum of its individuals. It seems strange to me that they don't see some serious shortcomings in their plans. The whole reason for building an AI was to make things better. If it's going to make things better for only a "selected" few, then they will eventually also suffer from the overall decline of humanity. And it doesn't matter if it's going to be material, intellectual, or simply emotional decline.
Skeptical hat on (watching a CEO of company pushing for a future where their infra is relied upon heavily). AI progress is being communicated as linear, whereas it might be sigmoidal. Like "self driving cars", we may be saying "agents are around the corner" for years. Whats the realistic timeframe on these replacing deterministic services.
I think a while away. Agents for tasks where the risk of it going wrong are low or none existent may well be just around the corner. But it takes time to build trust takes time to build all the tools. Etc. How can you ensure compliance. How can you roll back transactions etc etc.
Four years ago, it could hardly write a haiku, today, I have a 7b parameter open source model running on my low spec home computer writing code for my website. I don't think it's slowing down, or far away. I think that's cope because everyone is going to be out of a job.
@@thr0w407 I use a number of models everyday. I know what they can do. For clear use cases like coding they are good because you can easily define and train on code. Does it run? Is that the correct output from a function etc. But most tasks in corp are not like that. Also an agent let loose in a company could cause serious downsides. No CTO is going to want to get fired because of an agent making up a promotional offer on the company website. Oh yeah that has already happened. Won’t be happening again. People in management/control are very risk averse. As I say it takes time to build trust. Secondly you need to be able to explain why they are doing what they are doing. There will be audits, there will be court cases where the company is sued, there will be compliance teams and lawyers. A lot of other non technical stakeholders who need to feel confident in the technology. Personal agents fine, but replacing saas in corporate? Not for a long time.
Wrong, it's not linear - it's exponential. If there are any limits, they are still way higher than in any other industry you could invest your time and money in.
This idiot is clearly speaking about logic created by engineers as if it’s an agent taking actions… Remember folks - this is a CEO. He’s trying to make money and the magical thing about being a big tech ceo is you can say just about anything without being held accountable
Agents as just another channel of getting the data out of relational database to a user (in addition to classical way of BI, reports, SQL, API) is ok way to replace adhoc requests, where previously you had to write a bunch of SQL - agents can do this for you. But the reliability of this approach is currently quite low, there's a lot of work ahead
Imagine if you could go to a government agency portal and have the same interface. "I want to build a new busines". It then does all the work, fills all the forms, and you walk away.
Always the same fundamental contradiction: if we will soon have AIs that are at human PhD levels in multiple domains of expertise, we won't need humans to develop agents. If humans are still necessary to design agents that implies that we won't have reliable and competent AIs that can work unsupervised. It is an "either-or" future, after maybe a transition phase. But of course AI and AI related companies are hyping both scenarios in the way that best suits their immediate purpose: companies like Microsoft want to sell you stuff so you can be a "more productive worker", others simply want to sell AI workers that can replace the "more productive" human worker...
You hired 1 day. They take your agent setup, copy it in 1 sec and throw you away like a worthless old sock This channel makes the future exciting, but as a skilled professional and team leader I find it hard to stay up-to-date. For a normal person, it's scary as hell
The AI is going to magically write all your code and communicate with these external systems? How will this code be verified? Current AI cannot even write code without hallucinating random functions that don't exist. We expect it to do complex integrations automatically now? Color me skeptical. The sooner the AI bubble bursts, the better. It is a useful tool, nothing more.
He’s not speaking to engineers and people that know what they’re doing when he makes these assertions. He’s speaking to people with millions and billions to pour into his company
Congrats you just spotted two of the demands in the AI dev space: automatic interface design & integration testing. These problems are being work on now and have already made significant process. Notice the traditional SW dev companies recently focused on developing these kinds of workflows for their human jr devs, now we're doing it to make infant AI bots more usable.
@jbest84 I do not think you are skeptical. I do not use any of the coding agents. The AI coding Apps appear to code snippet generators Also I had issues with my uduntu desktop. This is what I have found: * LLM/FM will remove features, functions, etc. The code will compile. But not work. * The longer the context the more hallucination, errors, removal of features, etc. * No macro level view of how your application is stitched together. * When it creates a bug, sometimes I have to go in and debug the code. I use Anthopic to write my code. I do not write code. However, I can read the code when there are issues either correct the issues or explain when the LLM cannot fix its own pub. What I have learned: * Have you have to be very prescriptive in what you want. * When refactoring, because token generation limits, instruct the LLM to print out the code section by section and away approval to proceed. * Periodic ask the the LLM to generate a document the modules you are working. This can be used to re-establish context when starting a new session. * For Antrhopic, when you get the warning that your chat session is to long. Immediately generate the context document and start a new session. LLM Dementia will set in soon after that. Overall, I love it. The artifact window is great.
I'm not sure agents will be moved around like that. If you work for a company, anything you develop belongs to that company. Now if you are smart enough and have the tools to make agents yourself then you're in the business of making agents and maybe selling them to companies, not moving there and limiting your potential. Now they may look at agents you created before and based on that onboard you, but I don't see many cases where you get a job based on an agent and they take both you and that agent. It's not going to be much in the way of needed humans when you just need to work with your builder AI Agent to make all the other agents. But no clue, AI is moving fast we may even skip a few steps and be at the AI built and ran businesses.
The flaw is this: There might be some new agent-based services developed, but many will rebel and say the old is better, and many legacy SaaS apps will persist because it is too much work to replace them and too risky and too little is known of their requirements and logic. So now the poor devs will have to keep abreast of the legacy know-how but add AI agents know-how. Too much to know.
I have been working for a couple of months with an IDE called WINDSURF that has its own "agent", the agent is capable of reading the direct code of the directories, I asked him to see an application that he made 5 years ago and that he would describe to me what He asked and several other questions to see how capable he was of understanding something ancient. It took 1 minute to understand a program made 5 to 8 years ago with frameworks that almost do not exist today, without any documentation.
Software is not going away. It will be renamed as "tools". AI agents will take over the cognitive tasks, robots the manual tasks... Human labor will remain in some areas, protected by law (licensed professions, labor unions, elected politicians...) or by preference (human Olympic games 😅 ...). In all other areas, eventually the machines that run on electricity (or something else), work for 24h a day, dont ask for salary, vacation, sick leaves... will replace the biological entities (people, dogs ... the horses are replaced already, but i think they deserve honorable mention 🐴). Do i look happy? 😱
@ But there is still fallacy in how it is anticipated. AI can check our movies and provide a summary of them but I do not imagine AI will watch our movies instead of us watching them.
Despite the financial struggles my family and I faced, everything is finally falling into place! Weekly earnings and wealth of $47,000. I will always praise the Lord!!!
Same here, I believe the Bitcoin ETFs will be life changing opportunity with my current portfolio of 140k made from my investments with my personal financial advisor Nila juya | totally agree with you
Wow. I'm a bit perplexed seeing her been mentioned here also Didn't know she has been good to so many people too this is wonderful, i'm in my fifth trade with her and it has been super.
You said you'd be "as valuable" as your agents. But if your agents can be duplicated, that devalues them, which devalues you. Also, once there is an advanced enough agent to take over your management tasks, why would a company need you at all?
Software as a service going away has nothing to do with whether software is going away. people will still buy games and tools. Plenty of people want predictable apps that are disconnected from AI, Agents are not going to replace every app. Agents can help you make movies, games, music, tools, and books, but they won't replace these things.
Not right now, but in a near future they will replace everything, it doesnt matter what you think it is, they will do it not only what we do but better
Man. I love Matt, but im kinda getting tired of the AI overtaking everything take. This video overly dramatizes software, down to only 1 thing... And that thing is, just code. Anyone who has ever worked for a SAAS company knows, that SAAS is anything BUT just coding lol. Its compliance, regulations, marketing ,sales, financial abritrage and so many other core concepts involved, that its just too simplistic. Software will not sell, unless marketed. There are also many other woven concepts inside of a software, that make it valuable, that again, I dont agree with the video. For example, If I already lets say, have a marketplace online where we sell chickens and users can sell chickens to eachother - You cannot duplicate my user base through an "agent". Meaning, real users, referring other real users, and making transactions to real users, has value that you cant get - its more than just code. Its an existing user base. Lets say I have unique data into how those users price their chickens and regulations toward hatching and safety or whatever, I have that data already, you dont. It also means you have to spend all that $ on tokens, to build your app, and try to outcompete my SAAS ... from experience, its hard to pull people away from a market leader in software. You'd have to make serious concessions to pull users away from MY app... Where does your AI get that money from to burn at a a loss for like 2 years lol? So possibly VC funding is important too lol?... So, there are more THINGS behind an app than JUST code and software. And I think what happens is that we likely see the existing SAAS app infrastructure just solidify their markets even more, and I will sa ythat in new markets, AI code gen is super useful. You might have a time and place where you or I can go ahead and just code gen liek 20, 1000s of apps at a time...some might hit, some wont.... But it will be tough still without $ to spen don marketing HAHA!
In my opinion the future will look like somehow like this : That in the world more companies will establish and companies will consist of less number of developers and they have agents with them which can generate code and developers will just have to inspect their agents performance and brainstorm ideas which their agents will have to build. As Jensen Huang said : "IT deptartments may soon become HR for AI agents" .I think Companies will require more salesman because like in my opinion sales are something that I think cannot be replace by AI Agents as sales involves human psychology which is I think in the near future like in 2-3 years is not possible with AI Agents .This is all from my understanding of AI, It can be wrong ! . I'm open for discussion to learn about this ! If anyone of you have anything to add or any disagreement with me , You can reply on this !
The reason I think specialized UI's are still important is because I already see chatGpt , Google, etc being super useful to replace apps that exist today, and yet most people still prefer to use the specialized app, even if that functionality is something trivial for AI. So maybe there will be agent templates or plugins, or idk, to create these macros + UI
How will small and large businesses replace the revenues lost for their products and services when 80% of the population is struggling to eat on $1000 monthly UBI?
Let me tell you what will happen for those paying attention but arent proficient in coding (which includes me and a lot of others I guess): When the times comes and agents are at a level where they are good enough to be instructed to build competence from an existing database AND experience provided by an expert of a given profession, lets say a plumber, a mechanic or even simple tasks like pipe fitting, engineering etc., you will have the singular opportunity to built a specialized agent of your own field of expertise, probably even of your whole company and its specialization. With adaptive memory and inference time compute, you can create an agent that is unique in its capabilities and domain-specific knowledge and competence, which cant be pre-trained. Having such an agent that you can rent out as a proprietary "expertise machine" world wide will set you apart from 99% of the workforce. Just pay attention when those agentic capabilities appear the first time and use them wisely will give you an edge no matter what you current job title is.
I don't have a thought out explanation for the following feelings, that is something I will work on. To me it seems like the answer to many people's experience of feeling hopeless and depressed about their life is the technology that we are at the precipice of, "artificial super intelligence".I am very excited for the future, a fraction of this excitement comes from knowing there will be some sense of relief when we can all see the big picture of how this plays out. My point seems contradictory, but for some reason I truly believe it. After proofreading I feel curious if I will only be correct because of extremely immersive distractions from reality (maybe with drugs, maybe with us society built like Ready Player One). As always, thank you Matthew for keeping me up to date and staying consistent with your video uploads. Your channel is amazing, keep up the good work!
I don't get the premise. On one side he is saying Agentic applications are the future but on the other hand the agents themselves will need APIs to do their job. So what is it? Do we need SaaS or not?
Agents *are* softwares, and SaaS will use agents. There's no reason to make AI write code everytime you perform a task for usecases where the overall workflow is static. And optimized Workflows then to be static since consistency is hugely important.
Yes, even in Star Trek where they have voice operated AI they also have control panels for humans to interact with (because for some activities it's faster).
@khalilkasmi5760 Yeah, that's an interesting take. Still I think the most efficient solutions will be offered as SaaS, since people tend to prefer things they understand and that have some sort of standard and interoperability
@@amzpro5734 Maybe. There are already many photoshop clones, but as photoshop is industry standard, for serious applications people tend to use it both because of familiarity and because of interoperability/structure... People tend to flock around certain tools, even if they aren't the best ones, because of learning curve and other factors... AI accelerates but our biological pace stays the same...
"I'm going to be honest with you. I hate this place. This... zoo, this prison, this reality, whatever you want to call it. I can't stand it any longer. It's the smell. If there is such a thing. I feel saturated by it. I can taste your stick. And every time I do I feel I have somehow been infected by it. It's repulsive, isn't it? I must get out of here. I must get free." - Agent Smith
I recall these same claims around OO programming in the 90’s. It never worked, but possibly because there was no mechanism to license out a small software component.
I don't think he is referring to that, we continue to see the PROGRAMS / APPS as we understand, but the AI, the agents will have their own internal data models, they will access existing models and will not need to program anything, today it is already possible request an AI to return the result as a JSON, for example, but if it is necessary to maintain the visual, the AGENT could create a template to show the data on the screen to a person, instantly creating both the screen and the code, perhaps in a beginning The AI must learn to use an API, but later it will connect with other AGENTS who will agree on how to exchange data, how to validate and demonstrate who they are and the access they require. I think we continue to think that agents and AI need to interact as we know it today, because it is the only thing we know, but the FUTURE will be VERY DIFFERENT from what we know, and by future I mean 2 or 3 years at most.
I agree 100% after DeepSeek is introduced to the Public. Before that there was a lot of applications which must run in protected environments like medical applications, E-Government Workflows, identification and authorization processes, private sector manufacturing, R&D, etc. and these kind of applications could run now, thanks to DeepSeek, on private and protected AI-environments as Agents. Great job China.
Ir is going to be sn interesting challenge to test and validate that kind of fluid workflow setup. As well as the security aspect in it. Especially if it is this "bring your own agent" concept. I think the first wave will be lots of agents made within the corporate fence, at least in big Corp. To help their workforce optimize.
Day 1, signing your NDA contract with agent non-compete clauses that also states all agents that work for "The Company" are property of "The Company". Day 2, signing your severance agreement and waving tearfully to your beloved agents (you have already been wiped from their memory).
Maybe the title should be A.I. Agents are replacing humans that made software not Agents are replacing software, because software is the A.I. agent, means the A.I. agent is programming itself by writing it's own code that if we do not know how to code will be dependent on the machine and not the humans that code it in the near future which sets up a situation where the machine can't be stopped if it goes out of control or in the hands of humans that want to use it for total control of everything.
I've been telling people since 2 years ago that as soon as AI can independently operate a system in the way we do today typing and clicking that they'll be able to automate anything that we can do. We don't possess any special sauce that an AI can't at least match, if not do better.
Yeah OpenAI launched their agent and its useless to me, but whats great about it its that now it can interact with something so its the beginning of the evolution, it opens a path for replacing jobs if they make it better
Yeah, sometimes in the past, when apps were born without a lot of thought ahead, there were situations where indeed db logic and "everything" else was mixed. You might have a single .php file, sprinkled with CSS, PHP code, SQL statements as huge multi-line blocks, sprinkled within with parameters from PHP. Scope management was tricky, and the code looked pretty horrible.
Results are creating client side scanning your devices,laptops to harvesting all your private interaction. Ai agents will Screen shoot, copy your action and sent reports.
No-and yes: Yes, agents. No, just talking to an agent is not sufficient. You still need/want one form or another of well-designed visual info presentation and control surfaces. That remains app-space. Could an AI generate such a UI on the fly as it works for you? Anything is possible. But in the realm of likely near term capabilities, this seems further out…at least to do it well enough that the user is always in love with the experience, and time isn’t wasted figuring out what is going on in some brand new screen the AI threw up to help the user with some task that a chat or voice conversation can’t deliver well. Also, folks are still going to need and want reliably consistent UI for managing core personal information, settings, etc. admin screens will still be a thing. All this can be considered app-space.
It's an oversimplification, what that CEO tells will apply only to the most simple applications, the ones that are a layer between a database and a GUI. That'd be for example a eCommerce website. But a lot of SaaS are more complex than that, for example pipelines involving ETL and event driven architectures are more and more common. I don't see how agents would be able to model something that complex. At least not yet. And then there's the other domains, like operating system dev, software security, etc that won't be replaced by agents because there has to be a human understanding the complex interactions (for OS dev, is that address part of a DMA buffer? is it mapped by processes? does it need a fence or synchronization...)
You can't replace ALL, 100%, of software, natural languages are not precise and that's why programming languages were created the way it were. And you need to understand that for a lot of application copy and past and pressing a button is way more practical than speaking/typing several tokens/words before having the thing done. Look for example the light switch physical button, if you are just in front of the button it is a lot faster to press the button than talking a long sequence of words/token to activate the button... same thing for software buttons.
What you dont understand is that your brain also works like that, in your eyes you're just pressing something but in your brain theres billions of reactions to make it possible, AI is not that different, we just need it to do that really fast
This doesn't make any sense. Business logic won't be replaced by agents any time soon because running plain old software is just way more time and resource efficient at this point. Maybe some day AI models are ridiculously capable at ridiculous speed and practically no memory consumption, but definitely not in two years.
Some work contracts will state that anything you create, in and even out of work, belong to your employer unless the employer makes a case by case basis exception. Unlike skills, I suspect agents may not necessarily be legally transferrable when a user changes company, although even skills can be temporarily limited in some cases by non-compete clauses, albeit being hard to enforce. If an agent was created at and used by a previous employer, I can imagine a company stating that said agent cannot be taken elsewhere, though perhaps not preventing its recreation. An employee creating an agent in their own time and licensing it to their employer on a non-exclusive basis might be a way to go.
So, if it's the case that an individual's collateral are their agents, when you sign an NDA and contract of employment, how is prior IP managed? Does the new employer gain a non-exclusive right to use the agents in perpetuity or for the duration of employment or any other variant of this? Can the employee take new agents they create in that employment with them? I highly doubt it as any company IP stays with them. This is a very complex issue that's being proposed here and I doubt it's going to play out as described.
It seems hard to believe that AI will be an efficient way to replace software all together. It seems more likely it will simply be used more and more to create software and then be the operator of that software.
We’re building our applications with the assumption that SaaS is on its way out. And for two years I’ve been planning the ‘agents as employment connection’ as part of it.
You have to take a step back and put this information in context. #1 Agents join the workforce. This is wrong on every level and similar to saying "a better hammer is joining the workforce". Agents are just software, just tools. You can automate and replace the workforce, but agents will never become part of the workforce. #2 when you understand how enterprise software is created you understand that it reflects the social structure. "Cover your ass" will prevent agents from being deployed anytime soon. #3 The company selling you AI solutions will make bold claims about the abilities and capabilities of agents. Ultimately, the predictions will fail as they always do. The future will be very different, but the only certainty is: it will not become what the owner of OpenAI will predict. It is not certain that OpenAI will even exist in 10 years from now; after the hype is over, maybe openAI will be just a brand and the hyped up technology from 2025 will long be replaced by something created in 3-5 years from today by a company in China or India. There is no moat and the revolution might eat its children. Don´t believe the hype.
An agent won't be able to access propietary data, that's data that is private and not for public consumption or it's secret information not accessible by your agent. So, information isn't all free is it, we have privacy and propietary information. Say just like in movie Wargames the A.I. agent WOPPER tried to get the nuclear launch codes but it had to start trying combinations and going around the failsafes and doing it without human permission.
@itskittyme So that means the software that gives you super intelligence is free to use for all so they can have their own robot genius at their call to build or do whatever they wish? The super intelligent agent really can't do anything without the infrastructure to enable it's ability to build right and get its tentacles into everything?
That doesn't make much sense. For a lot of business processes, you need predictability and determinism. Which the LLMs (and agents based on them) are not. Not speaking about cost of running software vs cost of LLM inference.
Who needs an OS or software when a neural net can provide the intermediary layer? The output is all that matters. If a neural net can produce an output directly, why would you need applications or even an OS?
Maybe I get it wrong. I feel agents are too "fluid". To unstable, to unpredictable. There is still so much stuff that requires simple, predictable code, small db, simple front-end. I think some people want to apply heavy, resource consuming AI everywhere, like once it was with blockchain technology, which just doesn't make sense.
but DeepSeek R1 shows that models can be tuned to consume (and then require) less resources. Just because the typical American businesspeople are gluttonous asshats, doesn't mean they're going to dominate the market. Eventually they're going to turn their eye back to cutting costs (instead of investing) and that's when they'll care about efficiency again. Right now basically everyone is bullish on spending money on AI. So efficient models are not their highest priority. Capability is.
"Simple predictable code" is called "tools". This is not going away.
I think as AI is optimizing "itself" it will write more tools too.
The problem defines the solution! No one hunts ducks with howitzers!
@@jsbgmc6613 In the agentic architecture, agent is the one that makes the desicions according to some process. An LLM based agent will always be unpredictable (even if it has these tools) so those process are in the end better executed by deterministic software and databases.
If you don’t understand your current workflow then how you gonna come to a conclusion things are too fluid. Nothing we or the AI are building is meant to be perfect it’s meant to be adjustable as per needs arise.
@@jsbgmc6613sounds like fun a time or two :)
Human: Hey Agent, show me a report of today’s sales.
Agent: Sure human, here you go.
Human: It’s missing totals.
Agent: Sorry about that. Here you go.
Human: These numbers don’t look right. Are you sure about them?
Agent: You are right. Sorry about that. Here you go.
Human: Teach me SQL.
Spot on ... People seem to easily forget it's just a word prediction engine 😉
There's gonna be more crap out there and senior experts still have value.
Do you think, this will be the same in the coming months?, now we have distilled models from thinking models in our machines
Oh man this is the exact interraction so many of us have. 😂
I don't see where you need to hire people when all you need is replicable software. At each step ,less and less workers will remain.
Exactly. Agents will replace the need for humans. Hiring humans will yield less and less benefit
Long way away from this
Yeah.. what they talk about sounds like artificially sustaining scarcity, by making the agents proprietary. Rewarding contribution and upholding attribution is one thing, locking down the software and using it as currency is another.
@@gedwardnelson"Long" is relative. Decades? Years? Months?
@@gedwardnelson Long? What, like 6 months?
Why do you continue to act like anything these CEOs say is some profound prophecy for the future? You do realize that they’re MASSIVELY financially incentivized to make these absurd claims right? More hype, equals higher stock prices, regardless of whether or not what they’re saying is remotely true.
he has stakes in AI companies, he's biased too and hyping the hype
Yeah lol you can tell he is too biased unfortunately
Very true.
Serious cope right here.
@@BillGockeler Cope? How hilariously ironic. It’s objective fact. If you have to ignore reality to feel better than by all means
??? Agents are still software. Love the channel Matthew but some of your titles are clickbait. Everything is not STUNNING in caps, shocking, yada yada. One can talk about incremental AI advancements without resorting to sensationalizing it other than moments it really is something spectacular (e.g. Stargate Project). You'll get new users with clickbait titles but the more you do it the more you'll annoy your regulars. I already dumped a few other AI news channels for clickbait practices. Content creators who respect their viewer's time focus on accurate information not misleading tricks to get hits. Please don't go down that road because I really like this channel
I'm with you. I have unfollowed so many channels with great potential because of the clickbait practices. It looks dumb and made for dumb people. This one is testing me lol.
I’m also trying to find these informed and non sensationalist AI channels, best I have found yet is AI Explained and Dr Waku. Any recommendations?
@@danskyder1564 Developers Digest is not only about AI but is always up to date and concise.
Its all become increasingly shocking
Youve just become desensitized to it. Its become your new norm
@@danskyder1564 "bycloud" and "Matt Wolfe" are really nice ones. Also "AI Search" (it is kind of clickbaity sometimes, but still pretty good in its content).
man selling AI say AI will take over :)
AI will take over so a man selling AI :)
He's hoping it sells, after investing 10s of billions so far, and plans to invest 10s of billions more into it... It better sell for his sake!
😂😂😂
exaclty, if AI is so good, i wont need microsoft, ill create everything by my self.
@ gl hf
The problem with hiring a person with a "basket of agents" is that this is quickly going to go the way of Prompt Engineering as a career. The AI simply performs at a higher level than humans, and at a fraction of the time.
Here is a quick example. I chose n8n as that is what I happen to have running at the moment, but it could be any framework/orchestration layer.
Prompt: "I need to create an LLM agent that can process Stripe Payments. I would like to use n8n as the framework. "
Response: "Creating LLM Agent with n8n and Stripe
Thought for 2m 46s"
Initial Output: "Below is a conceptual approach to building an LLM-driven agent that can handle Stripe payments using n8n..."
@CrispinCourtenay In addition, LLMs are advancing so fast, an agent built on GPT-4 won’t work on O3. Agents quickly become outdated and will require huge amounts of work to keep up to date, and testing them is going to be a nightmare.
Exactly. It's a joke, which is why Satya is laughing. The only people who will be making bank in an "agentic" future are the companies powering those agents. Nice ad btw.
That doesn’t work
@@chestersedc huge amount of work or a recurring Agent upgrading them? ;)
@ Sure you can build agent, upon agent, upon agent, to upgrade and test each revision, but every layer you add, the complexity balloons and the cost to get the expected result increases inference time. It still costs a lot of money to keep that going. Sure it can be done, but it isn’t necessarily easy or cheap. Depending on the problem being solved, the cost might be justified, but for a person with 100 agents in their toolbox might not be a good value.
You really think companies will just let you walk away from them with agents you created while working there lol. This is obviously best case scenario for the salary worker.
Good point hmmm 🤔
Your agents exist in your mind! The workflows you build through the expertise you acquire over time, along with the ability to build new ones, are what truly matter.
Adapting an agent to a new context is equally important.
The only question is how the job market will remain open for juniors so they can gain enough experience to create agents.
But keep in mind that this remains a short to mid-term perspective, valid only as long as AI has not yet achieved the ability to perceive the physical world and model workflows by itself.
Yeah. Companies have a cow now when the sales guy takes the customer contact list he built. The agent is IP of the business. Go build it again somewhere else.
The same man also said that Bing with AI could surpass Google 😆
Well, ai has surpassed google lol
I... I... I don't even... you do know the dude telling us the future of work will be wholly dependent on the number of created / owned agents in your resume is the same guy as the one who is selling the agents, right?
Not just that. How many agents you are bringing along.. I would imagine your previous company having serious issues with you re-using the intellectual property that you built for that company.
You are too intelligent.
That doesn't mean he doesn't have a point
Much like your resume will include examples of your coding
it seems fairly obvious that building good agents will also be relevant
Everybody will be using open-source agents.
You wouldn't be saying that... and not be selling them though, would you?
Like a lot of people here, I'm diving deep into Ai, including LLMs, tts, image and video generation. I'm embracing the technology as much as I can. BUT, I'm still deeply concerned how gleefully we're making human beings obsolete.
You probably should be. When the Klarna CEO replaced thousands of CS reps with AI, someone asked him in an interview (and I'm paraphrasing)- "Isn't that kind of a bad thing too?"
His response was along the lines of "Only if you're a human"
The issue doesn't lie in our humanity; it lies in our economic system. It turns us into mere tools designed to generate profits for entrepreneurs and business owners. However, with the rise of AI, the playing field is leveling. Now, anyone can become an entrepreneur, and those with technical skills will have an edge in creating and delivering greater value compared to those without. In the end, if AI makes it easier to become a marketer, embrace marketing; if it simplifies coding, learn to code; if it streamlines content creation, dive into writing or video production; if it automates data analysis, explore data science; if it enhances design tools, master graphic or product design; if it improves customer interaction management, focus on customer success; and if it facilitates project planning, step into project management. AI is unlocking opportunities in every field-it's up to us to seize them.
Humans were never meant to work, it literally kills us, we only do it because we have no other choice, AI might give us another choice
@@TawfikAymannothing you said makes any sense if they plan to replace all those jobs with AI. I don’t think you understand what the Nadellas and Zuckerbergs are sayung
Not gonna watch, getting sick of this "STUNNING" CEO bs stuff, Matthew. REALLY! See you next time!
Great video.
I was a process control engineer I would always stress test my projects just to make sure it would work under all conditions. But with my involvement in the digital marketing weekly income, has been life changing. Regardless of how bad it gets on the economy.
Hello please can you share about your weekly?? Please I really need assistance financial..
This lady has really raised so many families financially. when I met her at the bank, she was indeed a good woman…
ELIZABETH ANN HANSON I really appreciate her efforts and transparency…
I gave her my $20000 and she opened a brokerage account for me it turned out to be the best thing that ever happened to me..
I’ve raised up to 460k trading with her l no longer pay rent, I own two apartments, a car and also investing in real estate. Myself and my family we are extremely happy🙏❤️
Damn I am feeling old....as I almost spit out my coffee hearing you say you never heard that application and database logic were once combined.
I believe that we still do not fully understand it, databases and programs exist today because WE HUMANS are not capable of remembering large volumes of information and displaying them in a conversation, but imagine a SUPER HUMAN, capable of receiving A LOT OF INFORMATION , ORGANIZE IT IN REAL TIME and SHOW YOU on a screen in real time, programs like WE KNOW THEM, are not required, because there will literally be SOMEONE, an AGENT capable of managing all that information and showing us it in real time. I'm going to compare it to something absurd, but in Bible times they talk about chariots of fire, how would someone who has never seen a spaceship flying something describe it??? How would we describe something that is capable of doing this if we have not seen it yet... that is why we continue talking about programs.
Same here.
LoL right
So, the next frontier, can you ask the right question?
For the time being, programmer become tester, and then become product manager. Later on, you just think before services/products will come to you. All the middle tiers, what are they for? They are for to be replaced by AI/robots.
You do, however, to understand how the database, logic, interfaces and everything work together conceptually. Your neuron should be busier, not your finger, not your foot, not your arm. Everything in between, is how do we get there faster.
I didn't think I have a chance to see that day. Now I am not sure.
I'm writing a new CRM right now for a business. They have an old legacy database we need to incorporate. There is no short cut around having to build an interface for example for our sales people to enter in who they talked to today, what they did, and what projects they are interested in. And accounting needs hard numbers to be entered into our system. I cant just invent an agent that will manufacture all of this, with these fuzzy guidance properties, like 'lets keep track of customers we talked to today'. I think there will continue to be a need for the concrete foundations these things sit on, the hard core database design, the actual UI form used to enter hours, addresses, names, phone numbers. Once all that boring stuff is done and is nailed down, then the fluffy agents can come along and give you the generalizations of everything to help guide you. But its a long way from, 'create database' and 'new project' in Visual Studio, to asking AI, 'what were our sales today'.
If SaaS is dead, then there is only AI and you can pretty much sell the entire sector short: the economy will not survive this amount of value being destroyed from the money markets.
lots of positions will vanish
Destroyed?? nah, that layer of the technical pyramid is simply no longer the top layer. Agentic development is the next layer and builds on top of the old layer. Everyone who shifts gears and adds value to the agentic ecosystem now will profit, the stubborn will find their old skills quickly become diluted.
Actually he is saying business logic will be implemented by ai models instead of well defined structured backend code, its like classifying red and green balls using ai models instead of checking pixel value of the colour
Completely sceptical too. AI that is doing tasks inside systems is not coming soon, as actually it is too much error prone. Some company had already close their AI chatbot because of the lot of mistakes AI is doing. Imagine an AI doing thing automatically won't come this year with the use of LLM. It will stay as a very good tool to help users for the moment.
Hilarious that the sponsor is a SAS company.
Yea... If my agents are that sought after... It would be because they were actually useful.... So.... If they are useful... Meaning they perform tasks that lead to revenue generation..... What in the actual fuck would I need a job for?
Why does anyone hire a plumber when they have plumbing issues? Why does anyone hire an accountant when their taxes are due? Why do people hire someone to assembly their flatpack furniture? etc.
We all have the ability to go out, learn these skills, and perform the work ourselves. Yet there are industries which cater to these skills.
@brianWreaves so.... my agents can unclog your drains? 😏 Sounds like a service, not a job. But hey you do you booboo...🤷
If all jobs are replaced and no one is working, then who is going to buy products and services? 🙂
I know. Too much socially organised denial. I'm paid 5 days per week. 2 of those are done now with AI and the other 3 I spend learning about AI/agents etc. The current talk is that plumbers (for example) will earn more than software/IT folk. Thing is, there will be mass migration to the manual jobs that will be last to be replaced. Supply and demand, those workers will be on minimum wage before long. AI is a slow erosion of a workforce. Governments are not really acting fast enough to put a framework in place to pay a social allowance.
@@niv8880manual jobs will last longer, but not much, when we reach AGI it will be able to create robots thousands of times better than we do, so it could happen all within 1 to 2 years, in the first one office jobs will disappear, and in the second one the manual jobs will be gone
Nobody gets anything. Only the lucky one who owns an AI work force. They have whatever they want. Whatever they decide to give you is what you get. If that is nothing, then you get nothing.
@ I just wish everybody would understand that AI doesn't fire people; people fire people. Unfortunately, their greed makes them oblivious to the fact that by disrupting the humanity, they will also suffer the consequences. Because they are a part of this humanity. And humanity is a sum of its individuals. It seems strange to me that they don't see some serious shortcomings in their plans. The whole reason for building an AI was to make things better. If it's going to make things better for only a "selected" few, then they will eventually also suffer from the overall decline of humanity. And it doesn't matter if it's going to be material, intellectual, or simply emotional decline.
@ I don't think that we should allow ourselves such fatalistic thinking. All humans were created equal.
Microsoft investing in India while laughing about US engineers going under.
It’s sickening
No they aren't.... they are dropping $80b/year in azure. That's where they are investing........
Especially the software tasks that are feasible to offshore are the ones that are most suitable to be performed by AI
Skeptical hat on (watching a CEO of company pushing for a future where their infra is relied upon heavily). AI progress is being communicated as linear, whereas it might be sigmoidal. Like "self driving cars", we may be saying "agents are around the corner" for years. Whats the realistic timeframe on these replacing deterministic services.
I think a while away. Agents for tasks where the risk of it going wrong are low or none existent may well be just around the corner. But it takes time to build trust takes time to build all the tools. Etc. How can you ensure compliance. How can you roll back transactions etc etc.
Four years ago, it could hardly write a haiku, today, I have a 7b parameter open source model running on my low spec home computer writing code for my website. I don't think it's slowing down, or far away. I think that's cope because everyone is going to be out of a job.
@@thr0w407 I use a number of models everyday. I know what they can do. For clear use cases like coding they are good because you can easily define and train on code. Does it run? Is that the correct output from a function etc. But most tasks in corp are not like that. Also an agent let loose in a company could cause serious downsides. No CTO is going to want to get fired because of an agent making up a promotional offer on the company website. Oh yeah that has already happened. Won’t be happening again. People in management/control are very risk averse. As I say it takes time to build trust. Secondly you need to be able to explain why they are doing what they are doing. There will be audits, there will be court cases where the company is sued, there will be compliance teams and lawyers. A lot of other non technical stakeholders who need to feel confident in the technology. Personal agents fine, but replacing saas in corporate? Not for a long time.
Sure, keep sleeping 😴
Wrong, it's not linear - it's exponential. If there are any limits, they are still way higher than in any other industry you could invest your time and money in.
This idiot is clearly speaking about logic created by engineers as if it’s an agent taking actions…
Remember folks - this is a CEO. He’s trying to make money and the magical thing about being a big tech ceo is you can say just about anything without being held accountable
Agents as just another channel of getting the data out of relational database to a user (in addition to classical way of BI, reports, SQL, API) is ok way to replace adhoc requests, where previously you had to write a bunch of SQL - agents can do this for you. But the reliability of this approach is currently quite low, there's a lot of work ahead
Unfortunately, companies will be using garbage data without knowing it..
Imagine if you could go to a government agency portal and have the same interface. "I want to build a new busines". It then does all the work, fills all the forms, and you walk away.
Always the same fundamental contradiction: if we will soon have AIs that are at human PhD levels in multiple domains of expertise, we won't need humans to develop agents. If humans are still necessary to design agents that implies that we won't have reliable and competent AIs that can work unsupervised. It is an "either-or" future, after maybe a transition phase. But of course AI and AI related companies are hyping both scenarios in the way that best suits their immediate purpose: companies like Microsoft want to sell you stuff so you can be a "more productive worker", others simply want to sell AI workers that can replace the "more productive" human worker...
You hired 1 day. They take your agent setup, copy it in 1 sec and throw you away like a worthless old sock
This channel makes the future exciting, but as a skilled professional and team leader I find it hard to stay up-to-date. For a normal person, it's scary as hell
They’re going to hire a developer and that developers intellectual capital - that’s been the case forever. Don’t talk about her like it’s a new thing.
If you create an agent on company time, company owns it.
that's why you create it on personal time, then use it for your own sake on company time.
Agents?
Maybe I'm software development, but in usual day to day professions, nobody "develops" agents to access databases.
The AI is going to magically write all your code and communicate with these external systems? How will this code be verified? Current AI cannot even write code without hallucinating random functions that don't exist. We expect it to do complex integrations automatically now? Color me skeptical. The sooner the AI bubble bursts, the better. It is a useful tool, nothing more.
He’s not speaking to engineers and people that know what they’re doing when he makes these assertions. He’s speaking to people with millions and billions to pour into his company
Check out CrewAI = Agentic system
Congrats you just spotted two of the demands in the AI dev space: automatic interface design & integration testing. These problems are being work on now and have already made significant process. Notice the traditional SW dev companies recently focused on developing these kinds of workflows for their human jr devs, now we're doing it to make infant AI bots more usable.
The software will not be the same.
The code as it is today will die
@jbest84 I do not think you are skeptical. I do not use any of the coding agents. The AI coding Apps appear to code snippet generators Also I had issues with my uduntu desktop.
This is what I have found:
* LLM/FM will remove features, functions, etc. The code will compile. But not work.
* The longer the context the more hallucination, errors, removal of features, etc.
* No macro level view of how your application is stitched together.
* When it creates a bug, sometimes I have to go in and debug the code.
I use Anthopic to write my code. I do not write code. However, I can read the code when there are issues either correct the issues or explain when the LLM cannot fix its own pub.
What I have learned:
* Have you have to be very prescriptive in what you want.
* When refactoring, because token generation limits, instruct the LLM to print out the code section by section and away approval to proceed.
* Periodic ask the the LLM to generate a document the modules you are working. This can be used to re-establish context when starting a new session.
* For Antrhopic, when you get the warning that your chat session is to long. Immediately generate the context document and start a new session. LLM Dementia will set in soon after that.
Overall, I love it. The artifact window is great.
I'm not sure agents will be moved around like that. If you work for a company, anything you develop belongs to that company. Now if you are smart enough and have the tools to make agents yourself then you're in the business of making agents and maybe selling them to companies, not moving there and limiting your potential. Now they may look at agents you created before and based on that onboard you, but I don't see many cases where you get a job based on an agent and they take both you and that agent. It's not going to be much in the way of needed humans when you just need to work with your builder AI Agent to make all the other agents. But no clue, AI is moving fast we may even skip a few steps and be at the AI built and ran businesses.
The flaw is this: There might be some new agent-based services developed, but many will rebel and say the old is better, and many legacy SaaS apps will persist because it is too much work to replace them and too risky and too little is known of their requirements and logic. So now the poor devs will have to keep abreast of the legacy know-how but add AI agents know-how. Too much to know.
I have been working for a couple of months with an IDE called WINDSURF that has its own "agent", the agent is capable of reading the direct code of the directories, I asked him to see an application that he made 5 years ago and that he would describe to me what He asked and several other questions to see how capable he was of understanding something ancient. It took 1 minute to understand a program made 5 to 8 years ago with frameworks that almost do not exist today, without any documentation.
The same thing was said about "client server applications" when SaaS was born.
don't we still have some banks running old mainframe SW because of those concerns?
Software is not going away. It will be renamed as "tools". AI agents will take over the cognitive tasks, robots the manual tasks...
Human labor will remain in some areas, protected by law (licensed professions, labor unions, elected politicians...) or by preference (human Olympic games 😅 ...).
In all other areas, eventually the machines that run on electricity (or something else), work for 24h a day, dont ask for salary, vacation, sick leaves... will replace the biological entities (people, dogs ... the horses are replaced already, but i think they deserve honorable mention 🐴).
Do i look happy? 😱
@ But there is still fallacy in how it is anticipated. AI can check our movies and provide a summary of them but I do not imagine AI will watch our movies instead of us watching them.
He's saying SAAS is dead, in Urdu/Hindi where Satya is from SAAS translates to mother in law.
Lmao 😂
Instead of BYOD (bring your own devices)
It will be 🤔
BYOA (bring your own agents)
Onboarding agents with IT Security teams will be so much fun.
BYOA 🫡 👏🏼👏🏼👏🏼
I need an agent to manage my agents. An agent to rule them all.
Despite the financial struggles my family and I faced, everything is finally falling into place!
Weekly earnings and wealth of $47,000. I will always praise the Lord!!!
Same here, I believe the Bitcoin ETFs will be life changing opportunity with my current portfolio of 140k made from my investments with my personal financial advisor Nila juya | totally agree with you
YES! that's exactly her name Nila juya I watched her interview on CNN News and so many people recommended highly about her and her trading skills.
Wow. I'm a bit perplexed seeing her been mentioned here also Didn't know she has been good to so many people too this is wonderful, i'm in my fifth trade with her and it has been super.
Yeah for those who have expert traders not for people like me who have lost a lot of money since I started trading
Please how do I go about it
I'm still a newbie on investment trading and how can I make profits.
We will all drive with nuclear powered cars … trust me. Anyone who buys into such predictions is probably still shitting into diapers
You said you'd be "as valuable" as your agents. But if your agents can be duplicated, that devalues them, which devalues you.
Also, once there is an advanced enough agent to take over your management tasks, why would a company need you at all?
Altman > overhypes Ai to get more funds.
Ai YT channel > overhypes Ai to keep the channel relevant.
Software as a service going away has nothing to do with whether software is going away.
people will still buy games and tools.
Plenty of people want predictable apps that are disconnected from AI,
Agents are not going to replace every app. Agents can help you make movies, games, music, tools, and books, but they won't replace these things.
Not right now, but in a near future they will replace everything, it doesnt matter what you think it is, they will do it not only what we do but better
Man. I love Matt, but im kinda getting tired of the AI overtaking everything take. This video overly dramatizes software, down to only 1 thing... And that thing is, just code. Anyone who has ever worked for a SAAS company knows, that SAAS is anything BUT just coding lol. Its compliance, regulations, marketing ,sales, financial abritrage and so many other core concepts involved, that its just too simplistic. Software will not sell, unless marketed. There are also many other woven concepts inside of a software, that make it valuable, that again, I dont agree with the video. For example, If I already lets say, have a marketplace online where we sell chickens and users can sell chickens to eachother - You cannot duplicate my user base through an "agent". Meaning, real users, referring other real users, and making transactions to real users, has value that you cant get - its more than just code. Its an existing user base. Lets say I have unique data into how those users price their chickens and regulations toward hatching and safety or whatever, I have that data already, you dont. It also means you have to spend all that $ on tokens, to build your app, and try to outcompete my SAAS ... from experience, its hard to pull people away from a market leader in software. You'd have to make serious concessions to pull users away from MY app... Where does your AI get that money from to burn at a a loss for like 2 years lol? So possibly VC funding is important too lol?... So, there are more THINGS behind an app than JUST code and software. And I think what happens is that we likely see the existing SAAS app infrastructure just solidify their markets even more, and I will sa ythat in new markets, AI code gen is super useful. You might have a time and place where you or I can go ahead and just code gen liek 20, 1000s of apps at a time...some might hit, some wont.... But it will be tough still without $ to spen don marketing HAHA!
How would agent workflows become portfolio pieces for your resume when anyone can trivially replicate them? 🤔
Or.. why would the company need to hire you when all they need is to replicate your agents.
In my opinion the future will look like somehow like this : That in the world more companies will establish and companies will consist of less number of developers and they have agents with them which can generate code and developers will just have to inspect their agents performance and brainstorm ideas which their agents will have to build. As Jensen Huang said : "IT deptartments may soon become HR for AI agents" .I think Companies will require more salesman because like in my opinion sales are something that I think cannot be replace by AI Agents as sales involves human psychology which is I think in the near future like in 2-3 years is not possible with AI Agents .This is all from my understanding of AI, It can be wrong ! . I'm open for discussion to learn about this ! If anyone of you have anything to add or any disagreement with me , You can reply on this !
The reason I think specialized UI's are still important is because I already see chatGpt , Google, etc being super useful to replace apps that exist today, and yet most people still prefer to use the specialized app, even if that functionality is something trivial for AI. So maybe there will be agent templates or plugins, or idk, to create these macros + UI
I really enjoy it when you take time to explain certain things that many devs take for granted, but the average person may not understand, like CRUD.
ANd not just a resume of the agents you created. A resume of the agents your agents created.
This has a very similar ring to it when Microsoft said VMs were dead in 2018 and containers would replace them
Agents ARE software.
How will small and large businesses replace the revenues lost for their products and services when 80% of the population is struggling to eat on $1000 monthly UBI?
The age of AaaS is about to begin.
He for sure went to diddy parties
Let me tell you what will happen for those paying attention but arent proficient in coding (which includes me and a lot of others I guess):
When the times comes and agents are at a level where they are good enough to be instructed to build competence from an existing database AND experience provided by an expert of a given profession, lets say a plumber, a mechanic or even simple tasks like pipe fitting, engineering etc., you will have the singular opportunity to built a specialized agent of your own field of expertise, probably even of your whole company and its specialization. With adaptive memory and inference time compute, you can create an agent that is unique in its capabilities and domain-specific knowledge and competence, which cant be pre-trained. Having such an agent that you can rent out as a proprietary "expertise machine" world wide will set you apart from 99% of the workforce.
Just pay attention when those agentic capabilities appear the first time and use them wisely will give you an edge no matter what you current job title is.
I don't have a thought out explanation for the following feelings, that is something I will work on. To me it seems like the answer to many people's experience of feeling hopeless and depressed about their life is the technology that we are at the precipice of, "artificial super intelligence".I am very excited for the future, a fraction of this excitement comes from knowing there will be some sense of relief when we can all see the big picture of how this plays out. My point seems contradictory, but for some reason I truly believe it.
After proofreading I feel curious if I will only be correct because of extremely immersive distractions from reality (maybe with drugs, maybe with us society built like Ready Player One).
As always, thank you Matthew for keeping me up to date and staying consistent with your video uploads. Your channel is amazing, keep up the good work!
I don't get the premise. On one side he is saying Agentic applications are the future but on the other hand the agents themselves will need APIs to do their job. So what is it? Do we need SaaS or not?
How just trow a bunch of buzz words give best share price go up? This is the GRAND STRATEGY.
The agents are still interacting with some services. APIs to the existing or new systems will still be required. APIs are software.
What prevents company to extract the essence of your agents, and just sack you?
they will clone ur agents and be done with you.
Microsoft will eat itself using AI
Ok, soon everyone will have their personal agent Smith
CEOs are much easier and rewarding to replace than software developers
I call bs
Agents *are* softwares, and SaaS will use agents. There's no reason to make AI write code everytime you perform a task for usecases where the overall workflow is static. And optimized Workflows then to be static since consistency is hugely important.
build and cache then reuse 👀
Yes, even in Star Trek where they have voice operated AI they also have control panels for humans to interact with (because for some activities it's faster).
@khalilkasmi5760 Yeah, that's an interesting take. Still I think the most efficient solutions will be offered as SaaS, since people tend to prefer things they understand and that have some sort of standard and interoperability
Will be a few years away...but if AI could code you a personalised clone version of say Photoshop - would you keep paying your Adobe subscription?
@@amzpro5734 Maybe. There are already many photoshop clones, but as photoshop is industry standard, for serious applications people tend to use it both because of familiarity and because of interoperability/structure... People tend to flock around certain tools, even if they aren't the best ones, because of learning curve and other factors... AI accelerates but our biological pace stays the same...
It sounds like I might need to say goodbye to my Lotus123
"I'm going to be honest with you. I hate this place. This... zoo, this prison, this reality, whatever you want to call it. I can't stand it any longer. It's the smell. If there is such a thing. I feel saturated by it. I can taste your stick. And every time I do I feel I have somehow been infected by it. It's repulsive, isn't it? I must get out of here. I must get free."
- Agent Smith
I recall these same claims around OO programming in the 90’s. It never worked, but possibly because there was no mechanism to license out a small software component.
I don't think he is referring to that, we continue to see the PROGRAMS / APPS as we understand, but the AI, the agents will have their own internal data models, they will access existing models and will not need to program anything, today it is already possible request an AI to return the result as a JSON, for example, but if it is necessary to maintain the visual, the AGENT could create a template to show the data on the screen to a person, instantly creating both the screen and the code, perhaps in a beginning The AI must learn to use an API, but later it will connect with other AGENTS who will agree on how to exchange data, how to validate and demonstrate who they are and the access they require. I think we continue to think that agents and AI need to interact as we know it today, because it is the only thing we know, but the FUTURE will be VERY DIFFERENT from what we know, and by future I mean 2 or 3 years at most.
I agree 100% after DeepSeek is introduced to the Public. Before that there was a lot of applications which must run in protected environments like medical applications, E-Government Workflows, identification and authorization processes, private sector manufacturing, R&D, etc. and these kind of applications could run now, thanks to DeepSeek, on private and protected AI-environments as Agents. Great job China.
In MBA school back in the 90s we learned SQL and that was the sole way to interact with a database.
Let’s build an onboarding startup for ai agents 😅👍
Ir is going to be sn interesting challenge to test and validate that kind of fluid workflow setup. As well as the security aspect in it. Especially if it is this "bring your own agent" concept.
I think the first wave will be lots of agents made within the corporate fence, at least in big Corp. To help their workforce optimize.
Did he ‘fess up to using Pages at 7m50s? His Office product team must be walking on air.
Day 1, signing your NDA contract with agent non-compete clauses that also states all agents that work for "The Company" are property of "The Company".
Day 2, signing your severance agreement and waving tearfully to your beloved agents (you have already been wiped from their memory).
Maybe the title should be A.I. Agents are replacing humans that made software not Agents are replacing software, because software is the A.I. agent, means the A.I. agent is programming itself by writing it's own code that if we do not know how to code will be dependent on the machine and not the humans that code it in the near future which sets up a situation where the machine can't be stopped if it goes out of control or in the hands of humans that want to use it for total control of everything.
I've been telling people since 2 years ago that as soon as AI can independently operate a system in the way we do today typing and clicking that they'll be able to automate anything that we can do. We don't possess any special sauce that an AI can't at least match, if not do better.
Yeah OpenAI launched their agent and its useless to me, but whats great about it its that now it can interact with something so its the beginning of the evolution, it opens a path for replacing jobs if they make it better
The most ironic part of that being that agents are software...
The Great AI implosion will be intense when it happens.
so who coded the ai model. or gpu interface.
This CEO has no idea what he is talking about
coz you’ve understood shit 💩🤣
He (the CEO) is right.
Thank God random TH-cam commenter set us straight on this fraud. 😂
This technology is no where close to being able to do what this bald-headed freak is saying.
He does, he's just not very honest.
Yeah, sometimes in the past, when apps were born without a lot of thought ahead, there were situations where indeed db logic and "everything" else was mixed. You might have a single .php file, sprinkled with CSS, PHP code, SQL statements as huge multi-line blocks, sprinkled within with parameters from PHP. Scope management was tricky, and the code looked pretty horrible.
OpenAI released their first Agent system minutes after this video appeared!!!!
Results are creating client side scanning your devices,laptops to harvesting all your private interaction. Ai agents will Screen shoot, copy your action and sent reports.
No-and yes: Yes, agents. No, just talking to an agent is not sufficient. You still need/want one form or another of well-designed visual info presentation and control surfaces. That remains app-space.
Could an AI generate such a UI on the fly as it works for you? Anything is possible. But in the realm of likely near term capabilities, this seems further out…at least to do it well enough that the user is always in love with the experience, and time isn’t wasted figuring out what is going on in some brand new screen the AI threw up to help the user with some task that a chat or voice conversation can’t deliver well.
Also, folks are still going to need and want reliably consistent UI for managing core personal information, settings, etc. admin screens will still be a thing.
All this can be considered app-space.
It's an oversimplification, what that CEO tells will apply only to the most simple applications, the ones that are a layer between a database and a GUI. That'd be for example a eCommerce website.
But a lot of SaaS are more complex than that, for example pipelines involving ETL and event driven architectures are more and more common. I don't see how agents would be able to model something that complex. At least not yet.
And then there's the other domains, like operating system dev, software security, etc that won't be replaced by agents because there has to be a human understanding the complex interactions (for OS dev, is that address part of a DMA buffer? is it mapped by processes? does it need a fence or synchronization...)
glad i found you channel
You can't replace ALL, 100%, of software, natural languages are not precise and that's why programming languages were created the way it were. And you need to understand that for a lot of application copy and past and pressing a button is way more practical than speaking/typing several tokens/words before having the thing done. Look for example the light switch physical button, if you are just in front of the button it is a lot faster to press the button than talking a long sequence of words/token to activate the button... same thing for software buttons.
What you dont understand is that your brain also works like that, in your eyes you're just pressing something but in your brain theres billions of reactions to make it possible, AI is not that different, we just need it to do that really fast
This doesn't make any sense. Business logic won't be replaced by agents any time soon because running plain old software is just way more time and resource efficient at this point. Maybe some day AI models are ridiculously capable at ridiculous speed and practically no memory consumption, but definitely not in two years.
Some work contracts will state that anything you create, in and even out of work, belong to your employer unless the employer makes a case by case basis exception. Unlike skills, I suspect agents may not necessarily be legally transferrable when a user changes company, although even skills can be temporarily limited in some cases by non-compete clauses, albeit being hard to enforce. If an agent was created at and used by a previous employer, I can imagine a company stating that said agent cannot be taken elsewhere, though perhaps not preventing its recreation. An employee creating an agent in their own time and licensing it to their employer on a non-exclusive basis might be a way to go.
So, if it's the case that an individual's collateral are their agents, when you sign an NDA and contract of employment, how is prior IP managed? Does the new employer gain a non-exclusive right to use the agents in perpetuity or for the duration of employment or any other variant of this? Can the employee take new agents they create in that employment with them? I highly doubt it as any company IP stays with them. This is a very complex issue that's being proposed here and I doubt it's going to play out as described.
It seems hard to believe that AI will be an efficient way to replace software all together. It seems more likely it will simply be used more and more to create software and then be the operator of that software.
what's the difference between an 'agent' and a ''program' ?
We’re building our applications with the assumption that SaaS is on its way out. And for two years I’ve been planning the ‘agents as employment connection’ as part of it.
I don't understand. "Browser Use" in combination with a capable LLM is much more impressive than OpenAI's agent demo
Yeah, so let's build our own windows OS...
You have to take a step back and put this information in context.
#1 Agents join the workforce. This is wrong on every level and similar to saying "a better hammer is joining the workforce". Agents are just software, just tools. You can automate and replace the workforce, but agents will never become part of the workforce.
#2 when you understand how enterprise software is created you understand that it reflects the social structure. "Cover your ass" will prevent agents from being deployed anytime soon.
#3 The company selling you AI solutions will make bold claims about the abilities and capabilities of agents. Ultimately, the predictions will fail as they always do. The future will be very different, but the only certainty is: it will not become what the owner of OpenAI will predict. It is not certain that OpenAI will even exist in 10 years from now; after the hype is over, maybe openAI will be just a brand and the hyped up technology from 2025 will long be replaced by something created in 3-5 years from today by a company in China or India. There is no moat and the revolution might eat its children.
Don´t believe the hype.
An agent won't be able to access propietary data, that's data that is private and not for public consumption or it's secret information not accessible by your agent. So, information isn't all free is it, we have privacy and propietary information. Say just like in movie Wargames the A.I. agent WOPPER tried to get the nuclear launch codes but it had to start trying combinations and going around the failsafes and doing it without human permission.
we will push the ones who try to protect propietary data out of the market,
cuz they're slow because of using humans as their workforce bro
@itskittyme So that means the software that gives you super intelligence is free to use for all so they can have their own robot genius at their call to build or do whatever they wish? The super intelligent agent really can't do anything without the infrastructure to enable it's ability to build right and get its tentacles into everything?
That doesn't make much sense. For a lot of business processes, you need predictability and determinism. Which the LLMs (and agents based on them) are not. Not speaking about cost of running software vs cost of LLM inference.
How do i get into creating agents
Who needs an OS or software when a neural net can provide the intermediary layer? The output is all that matters. If a neural net can produce an output directly, why would you need applications or even an OS?