[00:00:19] Happy to have you along, my fellow tech enthusiasts. I'm Sam Bushman, hard-hitting tech, always at your fingertips. Why is it hard-hitting? Because it costs companies lots and lots and lots of money, ladies and gentlemen, and that's hard-hitting to your wallet. I'll tell you that right now. Happy to have you along. Jay Hill's not with me today. He's at a conference for work. You know, we have to work too, you know.
[00:00:41] This broadcast is brought to you by NetworkProvidersInc.com. You know, you've got a strategic partner when you think network providers. You've got a partner that can take care of your endpoints, take care of your security, take care of your help desk needs, put up servers, build websites, and the list goes on and on. NetworkProvidersInc.com. Free. Consult to see what works best for you. But I'm telling you right now, you want to partner in the IT business, and we hope.
[00:01:10] That we make sense as your partner. NetworkProvidersInc.com. But I do have Ben Bushman with me. He works for a bunch of different people in the IT space. He works for NPI for one, and he manages servers, and he builds websites, and he's really a programmer, and does a phenomenal job building applications of all kinds. Welcome, Ben. Thank you, Sam. Appreciate it. Thank you, Sam.
[00:02:02] That reported on this claim that a company that he works for spent $500 million on Anthropic, and the report comes as businesses nationwide grapple with soaring AI costs. And they say growing token consumption is kind of the term they use. And businesses are racing to deploy AI tools. And now people are saying, wait a minute, are these tokens that we're spending, is this money we're putting into this really worth the expense?
[00:02:31] The incident is now being used as a warning that AI may require the same budgeting, imagine this, governance and oversight controls that's applied to cloud competing and other enterprise technologies. Now, to me, that should be a no-brainer. Why would you have to wonder that after this big, crazy expenditure? You would think that would be kind of from the get-go. Of course, we've got to regulate that. If it's not something that's a locked monthly bill, and I know you can say, well, Claudius, Sam,
[00:02:59] not for programmers and people that connect it for all kinds of systems and tools. And the more you use it, it's becoming un-metered. The meter is only for the chatbots, so the people that are on basic plans. If you want to connect that business-to-business, enterprise, system-to-system and everything else, and it's just going to run in the background, it's going to be a charge every token you use. The question becomes, how expensive does that get? Well, this company found out 500.
[00:03:30] This number is just, I don't know how to even think about this, Jay. I mean, Ben, what is it? 500 million dollars? 500 million on Claudius. I don't know how to respond to that, really. 500 million? How can that be? Can you really use that much? I mean, I wish I owned a company where I could spend 500 million on AI, and I'm just like, oh, man, that's a lot. But hey. Anyway, I find it fascinating because it just seems like it's just the numbers beyond imagination.
[00:03:59] I even didn't have it in front of me for a second there and thought, wait, am I right? Is it 500 million? 500 million dollars. Now, I don't think that's most companies. But I do think most companies better beware about their spend on AI. But Anthropic, by the way, announced a massive new funding round. It evaluates the company at a trillion dollars as enterprise adoption accelerates.
[00:04:27] Can you imagine a company that just started out relatively recently worth a trillion dollars, Ben? It's hard to believe. I don't think anybody can really understand the concept of a trillion dollars or a trillion anything. The only way I understand it is more than I have. I understand. Yeah, I was just talking to a lady at work, and she was saying, hey, I don't know what to do.
[00:04:51] When I'm writing things or doing things and prompting Claude, I run out of my session usage within minutes. Like, I get like 30 minutes of work done before, hey, I'm just – I can't work for five hours. And in a day of work, then, hey, I can't work with Claude for the next week.
[00:05:08] And it really goes to show that, you know, there's – you know, with Claude and AI, different agents, it's really going to take a large, you know, change in how we think about things. You can't just sit in a context window and chat it back and forth for, you know, minutes and minutes and minutes. You've got to create your context documents or your MD files, and you've got to, you know, start new chats to keep your tokens low.
[00:05:34] Yeah, and that's the problem with Claude and Anthropic is that it brings the whole string every time you want to do something. Other systems manage tokens a lot better than Claude does. One of the things that people need to learn to do is you just need to basically say, hey, give me a quick summary of our conversation now. And then you get that summary, and you cut and paste that somewhere. Then you leave, and then you come back and paste that summary, and it's, okay, here's where we were. Now I want to continue. And you've got to do that several times, and hopefully they'll get that eventually resolved. But that's the case if you want to save on tokens.
[00:06:04] That's one way to do it. The other way to do it is this. I know people want to pay $20, and they think that should be the sweet spot. And I would be critical of these companies that came out and started at $20 because they've kind of deceived people into that this stuff is really cheap. This stuff is not cheap. It's incredibly expensive. And so when you move to, say, $100 a month instead of $20, now most people can get work done. That's the real truth. And you would say, well, that's expensive. It's all get out. It all depends how much use you're getting out of it. What if you had an assistant? How much would that assistant cost?
[00:06:33] And how much are you getting out of, say, Claude compared to the assistant kind of a thing? I mean, that's a real, real discussion, Ben. And I want to talk about that because AI agents are now moving beyond chat. Digital workers are now writing code, researching the web, and reshaping office jobs as we know them. So when you think about an assistant that does all that it does, I mean, how many hours do you save, just say, a month from using these chatbots?
[00:07:03] If it rewrites a document for you, if you give it some ideas and it does this or that or it solves the problem for you, how many hours is it saving? And if you spend $100 a month on it, let's say you pay the bot $10 an hour, pretty cheap in America, right? Even minimum wage, just say $7.25 an hour. It's three hours a month, right, to justify the cost of the bot. Is it worth it, Ben?
[00:07:28] I think it's worth it for pretty much everybody, from HR reps to developers to even just business owners getting ideas for things. These things that are running constantly or just from your own prompts, it accelerates your work tremendously.
[00:07:51] The big question I have is, is it worth dropping employees or do you – how do I phrase this, right? Can I drop an employee and pay AI or can I just keep that employee, have AI for two employees, and accelerate my work so much that it justifies it? So I look at it and say, I think that you can honestly prevent or decrease how much you need to hire people.
[00:08:18] And I think if you've already got an overblooded staff, you can reduce a little bit of staff. But it's not this mass firing idea. Let's just say that you increase your Claude use or your ChatDBT use from $20 a month, so three hours of work equivalent, to $100 a month. Now it's 13 hours at minimum wage, okay? Or if it's $10 an hour, just say, then it's 10 hours. If it's $20 an hour, it's five hours only.
[00:08:42] And I would simply say this, giving everybody an assistant, if you add $100 to the cost of an employee and you give them this assistant, if they get more than five or 10 hours out of that assistant worth of savings and time and effort and whatever you want to say, if they're maximized to that degree, just look at it that way. Say we have now maximized every employee's output. So if I've got 20 employees and I was going to hire my 21st, I may not need to now. I might even be able to go to 18 employees.
[00:09:09] But you got to understand that not only is it going to benefit us from laborious tasks, not only is it going to increase the output of an employee, the other thing it's going to do that I think is really important for people to understand is it's going to let us work smarter. And so let's say that Ben's a 40-hour employee. And with Claude or whatever, Ben becomes worth, I don't know, the equivalent of a 60-hour employee. Say a 70-hour employee, whatever. We increase Ben quite a bit.
[00:09:38] And we have to pay $100 a month to get that done. That's nothing, right? If we turn Ben from 40 to 65 or 70 hours for 100 bucks a month, that's incredible. Even 200 bucks a month, that's fine. And so then I kind of come back and say with that advancement, how much of that increased productivity is fair to give to the company and how much is fair to give back to Ben? In other words, turning Ben from a 40-hour employee to a 70-hour employee for a minimum cost, minimum dollars.
[00:10:08] We've got to be honest as employers and say, I think we could work it out to where Ben can be worth, let's just say that he gets 70 hours. But let's say instead we can have Ben back off and he only has to work 35 hours. Now we give Ben a raise and his 35 hours is equivalent to 65 hours, whereas before his 40 hours was really putting in 50 hours. So now we back him off to 35, 40 hours. He's benefiting us by say 60 hours and we give Ben some time back too. I mean, isn't that the honest way to look at it, Ben?
[00:10:40] Probably. I've never thought about the concept before. But if you were a manager running a team, right? And you said, hey, my team is super efficient. You know, the company is going to reward that. And if you're an employee who says, hey, my work is super efficient, it should be rewarded. So 100 bucks a month times 52 weeks, just say five grand to make it simple. So I'm going to increase the cost to keep Ben as an employee five grand. How much of that time of increase, let's say it increases Ben's output 20 hours.
[00:11:10] How much of that can I give to Ben as a reward for being an learning AI, using AI effectively, maximizing his output with AI? How much can I give to Ben? And how much as a company do I need to take of that? That's kind of the fair, honest question. But artificial intelligence, folks, is rapidly evolving from, quote, simple chat bots into agents capable of performing real world tasks. Now, I want to talk about that when we get back with Ben a little bit more,
[00:11:39] because I want to divide this and help people kind of understand the landscape that we're talking about and what kind of productivity we can expect and what the gotchas are and everything else. You are watching the one and only Liberty Roundtable live. Cybercrime is exploding. Take Sarah from Sweet Delights, whose world crumbled after having to close with not being able to bounce back. Small businesses are prime targets,
[00:12:09] but the right strategies can keep yours safe. Jay Hill, CEO of Network Providers, has co-authored The Cyber Playbook Simplifies Cybersecurity for Business Owners with Strategies to Avoid Costly Breaches and Fines. Build a strong cyber attack response. Secure your business with key protections. Cyber threats aren't slowing down, but you can stay ahead. Protect your business. Ensure its security for tomorrow. Get The Cyber Playbook today
[00:12:39] at networkprovidersinc.com slash cyber dash playbook or call 385-446-5500 now. In the medical field, IT security is crucial. Our highly skilled consultants are HIPAA certified and have 20 plus years of experience servicing medical clinics, billing and supply companies. We offer comprehensive endpoint protection,
[00:13:08] guarding your computers and servers against all stages of threats. And with our 24-7 monitoring services, you'll never worry about extensive downtime again. Ready to level up your IT support? Call 801-706-6980 today and discover how great IT services can be with managed IT services.
[00:13:28] AI agents move beyond chat, ladies and gentlemen. Digital workers are now writing code, researching the web, and reshaping office jobs. Artificial intelligence folks, they say is advancing from just chatbots to autonomous agents capable of performing real-world tasks on behalf of users.
[00:13:57] Now, I find this fascinating because I want to divide this so people understand. AI does not do things. AI is nothing but intelligence. A chatbot then is an agent that acts and does things, but it's the automation or it's the application that does the work. AI just provides information. So think AI plus automation and now you're getting these agents that do things. If I ask a question, it's an application that gets me the answer.
[00:14:27] AI just produces the intelligence. If I now want to take action on what I've asked or what I've done or what I've told it, I've got to launch worker bees or launch agents. Those are automated. Those are, it's automation, not just AI. AI. So AI plus automation equals activity. Ben, do you want to kind of highlight this a little bit? I think that division is important for people to understand. Yeah. I mean,
[00:14:57] essentially, you know, the agent is either you or the automation, right? In AI. Yes. But I want to understand, people understand the AI is just the intelligence. The automation is the worker bees, the action. Correct. And that's why, that's why at first we had so much intelligence and so we could ask the bot and it would give us answers, but we couldn't automate tasks. And now it's rapidly evolving. So new data suggests that they're already being used far more than people think.
[00:15:27] Okay. According to tracking data from a new company called, it's an AI analytics startup. it's called arena. And arena says that the most common use of AI today, do you know what it is, Ben? Any guesses? My guess would probably just be research. Nope. The most common use of AI, listen, agents today is software development. Now that's why I wanted to kind of highlight this distinction.
[00:15:57] So when you come to AI, just asking it questions via the bot, you're right. It's research. It's writing up little things. It's doing that kind of stuff. That was the first kind of AI. And then they had a worker bee that could rewrite things for you. That was the first one that was kind of built in. But now separate from that, believe it or not, the most common use of an AI agent today, so an agent now is the automation side, is software development. Roughly 17% of actual automated agents
[00:16:27] are writing, testing, and debugging, editing, computer code. So 70% of the actual agents that are doing automation are code now. That's why the programmers are taking such a beating on one hand or so much more productive on the other hand, Ben. You're one of them, right? Yeah, I mean, really, we run agents to check for security issues, to look for bugs, to fix bugs, to do security audits,
[00:16:56] what can be improved. All that runs without us ever even touching it or thinking about it. Well, there is that. Those are things. But I also find that, you know, in the old days, programming was pretty simple. You had a group of libraries and you just wrote whatever you wrote and your computer was your computer. It might depend on an external thing like a printer or a keyboard or a mouse or something. But other than that, it was pretty much your computer. And then when the internet came and then the internet of things came and then all kinds of, you know,
[00:17:26] SaaS software and all kinds of internet online things started to come, the stack became much more complicated from a security point of view to it. And now you can't just know everything. So let's just say that you want to, you know, put a remote guest into your application. So, you know, like we're doing right now, we're connected and we're all remote from each other, but we're connected. How do you set up the link? How do you do all that stuff? How do you? And one of the biggest things I find is that you can lay out
[00:17:56] what you want to do. And the first thing you can get an agent to do is come back and tell you the scope of what you need for that project. It's tremendous. And so this is where I don't believe we should just fire all the programmers and think that AI is going to code everything for you tomorrow. AI can code, but it still takes the visionary. It still takes somebody when there's decisions to be made, who makes those decisions? Research now taps. 10% is research. So now you're getting bots to do research.
[00:18:26] And it used to just be ChatDBG would do the research, Ben. Now when I do research, I can say, you know what? Fan out. And when I tell it to fan out, it could get like 20 agents doing research for me. And they can divide the project up into 10 different, hey, Sam asked this complicated question. Here's 10 different topics I need to research. Let's have 10 agents go out and get the topics and then consolidate them and deliver them to Sam. That's what's starting to happen now. Well, when all this happens, I don't know if people understand this, but it's going to, the token use
[00:18:55] and the computing power use is going to skyrocket, Ben. Oh yeah. People don't realize what all goes into it unless you have understanding of computers or agents or machine learning or just code in general. You don't realize what it takes to power these things. And I mean, open AI or Claude Anthropic, right? Like Sam Altman said that, hey, it costs money to the $20 subscription to ChatGPT does not cover the cost to run it, right? They lose,
[00:19:25] I can't remember how much he said, hundreds of thousands of dollars a day, right? Running AI for everybody. And, you know, at some point that's going to have to shift, right? We're all dependent on it or getting close to it. And when it shifts, what's going to happen? Well, and that depends on how powerful it is and what kind of savings it provides. And that's why we need to educate the public about what it really takes. If you spend a hundred bucks at first, people are like, are you kidding me? I'm not spending a hundred bucks on a stupid chatbot. Well, it depends on what that thing can do for you,
[00:19:54] right? Unlike traditional chatbots that merely answer questions, AI agents literally take action. And so, it's one thing to say, how much am I going to pay for a Super Siri or a Super Alexa or whatever, just answering questions. The real pay dirt is in the work, the automation that it can do when it takes action. And what action can it take that makes it worth it for you? And if it can really take actions that make it worth it for you, and let's say that it costs,
[00:20:24] if you're going to, let's just say you pay an employee 15 bucks an hour. They're calling that now kind of a living wage, right? I mean, eight hours. Well, let's see, six hours, six times 15 is 90. So six, between six and seven hours, just say seven hours. Do you get seven hours of increased productivity? If you do, it makes sense to pay a hundred bucks a month for it. And so I think what we need to do is acclimate people into the reality of what it's doing, okay? These autonomous bots
[00:20:54] can browse the internet, gather information, create files, work with spreadsheets, interact with calendars, use other software tools to complete assignments that you give them. And you know what? Some people are saying, hey, these are digital employees that can be assigned work around the clock, Ben. And so people need to understand it's incredible and the dollars and cents are going to relate to that, but I want to let people know it's not perfect. And so then this,
[00:21:23] this group goes into the mistakes that it makes. Ben, do you want to talk about those a little bit? I mean, there are so many mistakes that AI makes, right? Like, you know, in my work, when I'm prompting for development, you know, I'll say, hey, you know, what do you think about this? And it'll give me a solution. Hey, here's your answer. This is how you fix it. And I'll say, is that really the industry standard? Oh, no, it's not. It's not the industry standard. Let's,
[00:21:53] let's fix this, right? Let's rework it. I've seen on LinkedIn posts about people saying, hey, we gave AI access to our whole, you know, repository of code and it deleted the entire thing. And we didn't, you know, it deleted our backups as well because it was running prompts and commands and things that, you know, they just gave it full access to their environment. And so, I mean, accidental damage, data loss, and just tons
[00:22:22] of unintended actions. So folks, there's a balance here and guardrails are essential and yes, it can do cool things for you. And yes, believe it or not, in most work scenarios, giving somebody an AI bot, most employees, it's worth a hundred bucks to give it to them. I kid you not. And that's the number when you really start to pay a hundred bucks, then you don't get battered by I'm locked down and I can't do this and I can't do that. You can actually work for a hundred bucks. It can actually work with you and for you and not just become a pain. But Arena, this is that company now,
[00:22:52] reports that agents falsely claim to have completed tasks roughly eight percent of the time. That's a massive number, Ben. In some cases, an agent may report that a file was created or a task was completed and it did not even happen. They say because many automated workflows depend on previous steps being completed correctly. And so then these, you know, cascading eight percents can become massive if you chain automations
[00:23:22] together, Ben. Well, not only that, but, you know, sometimes we think of tasks as small things, right? So if it fails eight percent of the time, oh, a hundred times gets it right 92 times. Well, sometimes those tasks aren't small. So those eight mistakes or eight times it fails, you know, add up to, you know, tons of work. And when you talk about chaining it, like you were saying, you know, the next piece then fails and the next piece fails because it was based off of false information that, you know,
[00:23:51] a task was completed or information was generated when it wasn't. And then a task to you and me, I'm sorry, go ahead. Hawkins said it just quickly falls apart and then you need a human to go in and recheck all of your agents and your workflows and make sure everything's set up. The other problem and I didn't mean to interrupt. The other problem is to define a task. So a task to me and you is not the same as a task to a computer. Let's just say this. I say this.
[00:24:20] Ben, go outside and, you know, clean off the windshield because it snowed today and we got to clean off the windshield so we can go. To you, the task is go clean off the windshield. When we come back, we ought to talk about what's the task to the computer, right? Yeah. I want you all to think about that. What is the task to the computer? There you go. All right. Thanks so much for being with us. NPITechguys.com Make it a great tech day, will you? Hey, thanks.