← Back to search
Is Middle Management the Real Job AI Replaces?
The Daily AI Show · 2026-10-08 · 57 min
Show full episode description
The episode opened with a debate over how much AI will actually displace human work. Microsoft AI chief Mustafa Suleyman highlighted economist Daron Acemoglu’s argument that AI may replace only about 5% of human work over the next decade, a sharp contrast with Suleyman’s earlier prediction that AI would reach human-level performance across most professional jobs within 12 to 18 months. The hosts questioned whether measuring complete job replacement misses the larger effect, since AI can absorb portions of several jobs and allow one person to cover work previously handled by multiple employees. Goldman Sachs offered a glimpse of what that shift could look like. New hires are already being described as managers of AI agents, entering the workforce with a “virtual army” rather than spending years performing lower-level analysis. That raises a different employment question: if entry-level workers start by managing agents, what happens to the middle managers who traditionally trained and supervised them? The conversation also explored whether companies could eventually judge human management potential by watching how employees interact with their agents. Jeff Bezos added another version of the AI productivity argument, suggesting some workers could eventually support a family while working three days a week. The hosts were skeptical, arguing that always-available agents may produce the opposite result. When an agent can keep working overnight if a human gives one more approval, the traditional boundary between work and personal time becomes harder to maintain. The second half moved into the agent ecosystem. Nous Research’s Hermes agent reportedly accounts for roughly 2.5% of global token usage, while Anthropic added monthly API credits to Max plans, released Haiku 5.5 at substantially lower prices and cut Sonnet 5.5 cache-read costs by 50%. Brian described already seeing lower costs in recurring agents running inside AJOVA Journeys. The episode closed with the coming competition between personal agents for scarce goods, from concert tickets to restaurants and campsites, ChatGPT generating New Yorker-style cartoons with real artists’ signatures, and Artcraft using Claude to build clean-room alternatives to Adobe products, including a Photoshop clone that reportedly handled real PSD files successfully. Key Points Discussed
✨ Episode Outline — click any point to jump to it in the episode
Problem solved
Whether AI displaces jobs outright or hollows out middle management as new hires become agent managers, plus the week's model and pricing news.
Benefits
- Goldman Sachs new hires manage AI agents from day one instead of years of grunt work
- Agent management rewards critical thinking over spreadsheet filling
- Haiku 5.5 makes cheap everyday agent work viable
- Anthropic adds API credits to Max plans
- Hermes lets users pick which models perform best for their work
Use cases
- MIT economist Daron Acemoglu's essay: AI replaces only ~5% of human work over 10 years, endorsed by Mustafa Suleyman
- Skydance CEO says AI will speed merging Paramount+ and HBO Max into one subscription
- Goldman Sachs associates deploy a 'virtual army' of prebuilt agents from first day
- Daily AI Show Spotify views grew from ~100 per episode to ~500 normal, 1,000+ peaks
KPIs / results
- Acemoglu: AI replaces ~5% of human work in 10 years
- Suleyman in May: human-level performance in most professional jobs within 12-18 months
- Typical span of control: 3-8 agents per worker
- Show Spotify: ~500 views normal, sometimes 1,000+
📑 Chapters — tap a time to jump there
00:02:47
Will AI Replace Only 5% Of Human Work?
00:06:17
Why The 5% Number May Be Misleading
00:09:08
AI Accelerates The Paramount-Warner Integration
00:12:17
Goldman Sachs New Hires Manage AI Agents
00:14:42
What Happens To Middle Management?
00:16:13
Gareth Joins The Conversation
00:17:51
Could Companies Judge You By How You Manage Agents?
00:20:23
Jeff Bezos Predicts A Three-Day Workweek
00:22:19
Could AI Actually Make Us Work More?
00:25:01
Are Workers Any Less Stressed With AI?
00:28:00
Why Humans Still Cannot Truly Multitask
00:29:16
Dot Shows A Different Model For Personal Agents
00:34:39
Anthropic Adds API Credits To Max Plans
00:35:15
Haiku 5.5 And Lower Anthropic Pricing
00:37:15
What Happens When Everyone Has A Ticket-Buying Agent?
00:41:30
Can Lotteries Solve Machine-Speed Queues?
00:47:03
ChatGPT Signs AI Cartoons With Real Artists’ Names
00:48:49
Claude Helps Build Open Alternatives To Adobe
00:52:20
Can Haiku 5.5 Handle Cheap Everyday Agent Work?
00:55:25
Which Models Perform Best With Hermes?
00:57:19
Episode Wrap-Up The Daily AI Show Co Hosts: Brian Maucere, Andy Halliday, Beth Lyons, Gareth Hood.
Hey everybody, what's going on? Welcome to The Daily AI Show. Today is October 8th, 2026 and I'm joined by Andy and Beth, Brian. Glad to see you all here. We're here every single weekday at 10am Eastern so make sure you come back and hang out with us if this happens to be the first time you found us, which happens every day. So we're always happy to see our new folks. If you want to join us live, the easiest way to do that is through YouTube. Like I said, we're 10am Eastern so if you happen to be around during that hour or happen to have a day off, come join us live. Be part of the community, be part of the folks that are adding to the show while we're actually doing it. You'll often hear us talk about or bring comments up on screen. So if you want to be part of the show in that way, you can. But if you don't have time during the 10 o'clock hour, that's certainly understandable. Another great way to do that is to go through our community. So you can go to thedailyaishowcommunity.com and over there you can just sign up. It's a free Slack community and it's a great way for you to join the conversation. More asynchronously on that side of things. So there you go. There's all the ways that, Hey, if you like what we're doing, a like or subscribe along the way, uh, goes a really, really long way for us. It just helps us reach more people. I was just sharing with the team the other day, what our September numbers were for Spotify. And, uh, yeah, we continue to, to grow at a, at a real, really like consistent pace with, with some, you know, speed ups along the way on there. So, yeah, I, I can remember on Spotify being really, really happy. If we bounced over a hundred, um, at that, they, they've changed the metrics just like YouTube does. So it's not always the same, but it was a hundred and meant like a hundred views, right. Or a hundred listens or whatever on Spotify. It was a hundred. And if I saw like, if we were at 89 and like, we had a show where we popped over like one Oh one or whatever, I got really excited these days. If it's not, if it's under 500, I'm like, Ooh, I wonder how we missed on that show. But usually it's the title or something. So it just goes to show you can get used to anything. You can get used to any numbers, but I do try to remind myself that once upon a time, when we started this show, seeing anything over a hundred was really great. And nowadays if it's 500, it's a normal show. And sometimes it can bump to a thousand or more just depends on what the topic was and everything else. So, uh, anyway, thanks everybody for being here. Let's dive into it. I know we have some, uh, new news coming out about some of the, um, the models and things like that. So I wasn't here yesterday. So with that, I'm going to kick it over to you guys. Cause I'm not quite sure what you guys spoke about yesterday. I'm going to talk a little bit here about jobs and, and, you know, the up and down question of whether AI is a job displace or not, because yesterday, uh, the head of AI at Microsoft, Mustafa Suleiman, the person behind Pi, the original emotional, uh, AI, uh, tuned thing, uh, which was acquired by Microsoft really to acquire Mustafa Suleiman. Uh, and, and I think he has a, uh, uh, either his, uh, bride or a partner or somebody else was also done in that acqui-hire. Anyway, yesterday, uh, he wrote on X, uh, a post that kind of shifted his perspective on this toward the endorsement of and promotion of a, a recent, um, essay that was written by Darren Asimoglu, who is an MIT economist. And who's widely regarded as a smart guy in the world of economy, but the, the essay, uh, basically says that AI will replace only about 5% of what humans do over the next 10 years. Now that's very different from what the, what it feels like, uh, you know, when you see so many people in leadership positions inside corporations saying, okay, we want to automate things and we want to reduce our head count. And that actually happens, right? And we have plenty of numbers associated with that head count reduction, particularly in areas where AI has a serious contribution to knowledge work. And so a lot of people are worried that AI will take their jobs and there's a resistance that builds in the workforce against AI because of that. And Suleiman is shifting because just in May, he said that AI would reach human level performance in most professional jobs within 12 to 18 months. So that just empowers the notion that, Hey, all of these knowledge work positions can be replaced by AI. And increasingly we are now treating AI agents as colleagues in the, in the team who can actually be delegated to. And I have another note that I want to talk about, which has to do with Goldman Sachs. And I'll come to that in a moment, but anyway, uh, Darren Asimoglu, the economist says that, uh, AI developers should stop building AI. Automations and, and, uh, agents to replace people, but rather shift their focus instead on making humans more effective using AI. And that makes sense, of course. Um, and, and, and Suleiman basically points out that, you know, AI isn't showing up in the productivity statistics yet. And that's one of the things that supports his shift towards Asimoglu's, uh, you know, thesis, which is that, Hey, AI doesn't have the ability to replace humans, but in co-intelligence between human and AI, there is this slower increase in productivity. Productivity and GDP, you know, that's traceable to AI, but it's not, it's going to spread out over 10 years. It's not a revolution. Can I, can I talk on that a little bit? Cause I think, okay, here's, here's some of my thoughts on that. Number one, 5% is still, yes. It may be not as 95% or, you know, like the big, big numbers, 5% is still a massive, a massive impact on a global workforce. Right. So that that's still, even if it was 5%, it is 5%. But what I, what I think we all see, we all know this. We talk about it all the time is, you know, if the goal, if the, if the 5% is measured against 100% and only 100% of replacing a job, that's one thing. That's a very hard number to hit, right? That's like, you have to replace all of Brian, everything, everything I do. It has to, for, for that 5% to be whatever, to say that number, I'm assuming that's what they mean. It literally replaced me. The company removed Brian and replaced him with some sort of AI solution. And Brian wasn't best, right? All my, that, that's not really, it's not that clean, right? It's a lot messier than that. So I agree with what you're saying as far as like it being about empowering and making the human work worker more efficient, which is absolutely what we're seeing with, you know, if every worker had 20, 100, 1000, you know, agents, not that many, but you know what I mean? A span of control is usually three to eight. So let's just start there. Let's just say every human, every, you know, a worker in a, in a, um, SAS company, let's just keep it niche or whatever had three to eight agents also working for them. That still means you're going to have a reduction in workforce for sure. Because, okay. It took, it took 40% of my work, but it also took 20% of Beth and it took 15% of Andy. And then the company goes, well, we only need one. We just need Beth. And Beth can take Brian's 40 and Andy's 15, and she's still capable of doing everything she needs to do for the job. Brian and Andy are dispensable, but for whatever reason, right? So you're still going to see AI have a massive impact on reduction of workforce. I think in certain sectors, not everywhere. And also I think that 5% is probably misleading because if the true criteria of that percentage is literally replace that person, as in they are not missed. There's no reduction in, in capacity or ROI or anything, whatever metrics you're looking at for that particular person of what they did for their hundred K, you know, uh, you know, salary or whatever. That's, that's different. And also I don't have to pay healthcare to agents. I don't have to pay either. So, so there's a lot of savings over there too, when the reduction is from, you know, to Beth. So Beth is working just as much as she ever was, but now she's got three to eight agents also working with her, which have absorbed the work that I was doing and Andy was doing. And you know what I mean? Like, so I just think it's, it's a lot more messy than perhaps some of these statistics are doing. And then I'll just throw in one news story as a, an example of this, uh, the merger of, is it Paramount and Warner brothers, that big merger? Well, the, the new CEO of, I believe it's Skydance, which has been around, but that's also going to be the new name of the company. I believe if I have that right, the big, the parent company is Skydance. It's going to be, it's going to be Warner brothers, a Skydance company. Yeah. Ace Skydance company. Okay. So he was saying in, in something that, uh, Oh, even the future, uh, Paramount plus and HBO max will be merged. They'll be brought into one subscription service. Now we've, we've heard and seen this Disney still trying to merge in Hulu, right? And who's going to go away. It's just going to be Disney plus they've been saying this for a long time, but it's difficult. But the point I want to bring up is he said, I think we're actually going to get there faster. We'll have bundle deals. Sure. That'll come up pretty quick, but because of the advances in AI. I think we're going to be able to do this merger of these two subscription services much faster. Now that's just a throwaway line that he gave while talking about else, but what does that really mean? That means people are probably not as many people are going to be employed on that merger. Well, you know, even before the advent of AI consolidation, which is what happens when a company merges has always meant that there are going to be efficiencies in the combination that justify the expense of the merger. And that typically means getting rid of duplicate management layers that happen in both of those independent companies as you combine them into one. But now AI is actually accelerating that process so that a merger and a concentration, a monopolization, as is the case with this particular one in studios, is going to concentrate in the hands of a much smaller group of people. The production pipelines for media. Yeah. You're on mute now. This gets to the place that I sort of come to from this because what used to be state of the art was that your critical capabilities needed a redundancy. Right? I go out. I'm in the hospital for a month. Brian actually knows enough to be able to run what I did so everything doesn't come to a halt. And I don't see that level of backing in managing agents. Right? And I feel like what is happening is that the agent management is faster but more intensive and personalized for the person managing them. And not necessarily like you can't bring in somebody else and go, how do I manage you? Right? Like because now we've taken an active manager who helped design the system to someone who's like, you're just going to tell me. Right? And that's the place where efficiency becomes a big flaw, big problem. Here, I have a weave on that, which is at Goldman Sachs. This is also in the news in the last couple of days. At Goldman Sachs, they're saying that their new hires coming in are acting as managers now of AI agents. From their first day, their job is no longer to do the analysis and learn over the course of several years. They're put in charge of agents that have already been built. And they have to manage those agents. Because as you point out, Beth, it takes a lot of time and attention and knowledge in order to manage agents. And it's a managerial task as opposed to a sort of low-level analytical execution task. And learning those skills can be done by these. Remember Goldman Sachs, if you get hired on at Goldman Sachs as an associate, you've probably done a serious academic performance already and you've been recruited into one of the most elite positions. But now your job is really about doing that. And so what one of the leaders at Goldman Sachs said is, when our young folks now start work, they're managing agents. They have to make the most of this virtual army that they've got. And in earlier generations, of course, bankers had to work for years in order to learn the skills. But now the agents carry with them enormous context and knowledge about that. And the job of the new hires there is to just deploy all of those agents. Which means that if you're coming into those jobs, people are being recruited. The ones are going to be able to demonstrate through portfolios or internships, critical thinking skills. Because what do you have to have if you're going to manage agents? Critical thinking skills. Because you're not being asked to fill in the spreadsheet. They're saying, no, no, no. We have a technology that will fill in the spreadsheet. Don't worry about that. So like, whereas before it was come in, we think you're really intelligent. We think you're going to be great for Goldman Sachs. You're going to start at the bottom and you're going to do some paper pushing. And that's exactly where you should be because, you know, you can't hurt anything and blah, blah, blah, blah. Well, why can't I just onboard me now managing agents? But the whole point of that is that I am maybe working with some sandbox data or some live data or whatever. And I'm able to figure out what my critical thinking skills are and actually learn the business probably much, much faster. Potentially, depending on how they do it, safer as well. Yeah, but the implication of that is if the new hires are managers now, what happens to the managers who used to manage the new hires? Right. So there's the threat is to the middle management layer that that was largely responsible for the sort of the corporate knowledge transference and the training and development of the new generations of people who would rise into middle management. But now I don't call them a manager. I think that's really you wouldn't call a new hire a manager just because it would be just because. Well, but Goldman Sachs is calling. They're saying, look, they're hiring and calling them managers. No, no, no. I'm not saying what their title is. I'm just saying their quote is that these people that we're hiring in are managers of agents. That's what their job is. In the future, is everybody a manager of an agent? Yeah. Which is different in the news coming out of banking in terms of employment than the news that was coming out of tech in terms of software engineering. Software engineering was we're not hiring the entry level. We can get it done with the people we already have. In banking, we're letting go the middle level potentially. And we're happy to hire more AI native people who we would have had to train in analysis. And now they're in a position to be able to do this kind of work from a management standpoint. And that's the skill. And Gareth is with us. Yeah. Also raised his hand, which will get you nowhere on this show, Gareth. I know. You've been here long enough to know that if you don't jump in, you got no shot. I see talking over people and I feel rude. Nope. Not on this show. And there's another... We'll learn that out of you, buddy. Yeah. There's another aspect of it is like some managers are people people. Like they're really good at talking to people. But this kind of... And that's like really what drives them is being able to manage people. And they're really good at managing people, right? And now the workforce is going to change away from that. And so I worry like when I was in banking, I had some great managers who were really good at their job because they were really good at working with people. But no longer do they have to work with people anymore. So that like really changes that, those type of people. And now I wonder, like I never thought about this. Like I wonder what happens to that side of the workforce because just because you're good at working with people doesn't mean you're going to be good at working or managing agents. It's a different skill set. So where do those group of people go? What happens to those people? You're still going to need them. I mean, look, you're not going to not need people people, right? It's like that's a really, really good skill. And you're right. Not everybody has it. Conversely, I will also say that if you on that same track, if I could see a future where somebody... Let's say Goldman Sachs. I'm not saying this. I have no idea if this would happen. This is hypothetical. But let's take the Goldman Sachs manager of agents. Okay. Six months goes by. They say, Gareth, you're doing really... We really think you're doing really great work. And you're like, yeah, I'd be curious what my promotion potential is to a manager of people. Also, those people have agents and I have my own agents. Okay. We agree. Right? And they go, yeah. Here's the thing though. We can see all of your communications clearly. We've told you we would of you with your agents and your interpersonal skills with agents does not really give us the warm and fuzzies that you will be a great manager of people. And so you're actually not on our fast track. We actually prefer people who are polite, who understand how to reason, don't get upset quickly. But I see peace every time. I see this happening, man. If anything you're talking to an agent about at your work is the property most likely of that company and it could be probably reviewed. And we were just talking about... I wasn't here yesterday, but two days ago, you guys were saying how much you do voice activated stuff. Well, what if all the AI tracking me talking to the AI is being like, Brian gets frazzled quickly, man. And then gets upset at the agent because it didn't understand. If that's the way he's going to handle an agent, we're just going to assume that's his people skills. And maybe he's not the fit for that. We're going to take people who don't get frazzled and actually manage their agents in the same human way they would do it. It's the Carl kicking the robot who can't take out the recycle. Some people will say it's a robot and don't kick it. Other people are going to say it's a robot and who gives a you-know-what. It's just trying to take out the thing. That is part of when you apply for a job. Now it's going to be, do you kick a robot? But, man, but like, yeah, I mean, I'm not- How cute does a robot have to be before it's not okay, you know? I'm not good at managing people. When I worked in retail, I ran, I mean, I was a store manager for like more than a decade in retail. And I learned eventually that I suck at managing people. I just don't, I'm not good at it. Some people know that. Yeah. I'm not good at it. I'm too blunt. I'm just straight too blunt. And that's why I'm good at managing agents, I guess. And so, yeah, I mean, but it's interesting, but there's also people out there that are really good at it. So, yeah, it'll be interesting to see what happens. So here, I want to follow on this same thread with something that yesterday Jeff Bezos said in a Fox News interview. Okay. So he got on Fox News with Brett Baer and he smugly said that artificial intelligence could make workers productive enough that some of them will be able to support a family while working only three days a week. And he also pushed back on warnings that AI will cause widespread job losses. He said, you know what? Some people will decide I can support my family just working three days a week. No, wait a second. Like, and the company's going to say, okay, no, no problem. You just work three days a week. We're going to pay you the same amount. Right. Yeah. That, that's a little bit of a question there. Amazon specifically. Right? Yeah. Yeah. Is Amazon going to go for that? Yeah. I can support my family while having two bathroom breaks in the middle of a day. Like, Amazon is not. Now, he further said that in some two income households, one partner might decide to stop working altogether, but he doesn't explain how AI would make a three day week possible or allow for only one person to make enough money to supplant the loss of the other person's income. But, but he is while spouting all of this, he's making it much easier for people to make that decision to lose their job because Amazon cut finalizing in January, 30,000 corporate workers. Okay. So, you know, here's on the one hand he's saying, oh, you know, this is going to be a, you know, a nirvana. You're all going to be able to work less and make as much or maybe more at the same time people are unemployed. Just. Yeah. Which is look, that's a fallacy. We all, I think we all maybe thought that might happen, but we all know what's really happened. Now I have to manage agents. Now I have to manage, I have to manage my three to eight agents, my spanner control or whatever. That doesn't make me work less. We've all said this on this show. Geez, I think I'm working more these days than ever before because if you, they're always on easy to get to me. Oh, Brian, just check your phone really quick. And, you know, like, you know, just tell the agent really quick what it can do so we can just keep moving. You wouldn't want it to not work overnight. Right. And then you go, yeah, but I don't know, 6 PM. I don't really want to go. Yeah. But if you just give me these three instructions, I will take it from here, Brian. I'll take it. And all of a sudden I'm like, Jesus Christ, I'm on my phone at 7 PM at night answering these dumb agent answers because that's exactly what's going to happen. And then it's going to be the weekend. And then you're going to be like, geez, I used to be done with my job at five or whatever. My nine to five is not a nine to five anymore because it's got me connected and it's making me feel guilty that if I don't give it this next approval, then I'm really hurting the company because it would be happily more efficient in my absence. But if I don't get past the blocker of the human, you know, that's a problem. So this whole three day work week thing. Yeah. Maybe if you're contract, there's nothing wrong with that, Andy. Right. I mean, you can, you can decide to go contract and know what your builder and maybe you, you have multiple, you know, eggs in multiple baskets through whatever you're doing. And yeah, you can work three days a week that way because you make enough to support your family. Yeah. But a W2 job, I don't see it. Yeah. I just want to make a comment about contract work or consulting work. There's an enormous administrative overhead to managing that kind of responsibility for your own income streams across multiple clients. That takes a ton of time in itself. And it used to be like, go all the way back to the idea in Japan of lifetime employment. You know, you joined a company and you basically wore the pin that's the flag of that company for your entire life. And, and, and they took, gave you a pension and you retired with honor and all of that. That's just not happening anymore. You are expendable. You are a cog in the wheel at this point. Of course. And if you, if you go away from a W2 job and try to take responsibility for managing a level of income in three days a week. Now it's, people are doing it and, and the, you know, they, they post on Instagram or on TikTok and show, Hey, I'm, I'm over in the Bahamas right now. And, and I'm making tons of money and I, and I'm, you know, juggling all these things. Well, it does happen, but it's not a solution to the general workforce out there. Do you think, okay, so people are, maybe we're working more, but do you think people are less stressed? No. Oh, I think it's, I think it's very stressful to manage agents. And it, and it's, it's also, unfortunately, and I don't know how to solve this part, but I sat on my ass in that chair over there for nine hours yesterday. Yeah. No, that, that's not good for me. Yeah. So yeah. Anyway. Yeah. But you could also get up and talk to your agent on a walk if you wanted to. Yeah. I'm getting there slowly. I'm trying to get my, my tool set together to do that. But here, but, but this is, this is exactly my point though, Andy, you will get there. Right. And you'll be like, Oh, isn't it great that I'm still making progress on this. I'm a nonprofit and the other things that you're working on, right. Your side projects or whatever, but I'm out with the horses. Hey, I had to be out with the horses anyway. This is great. But then you go, geez, man, now I'm stressed while I'm working with the horses. And that used to be something that was manual labor, but it was a release. So, so how, even when the market's Andy had to solve the problem. This thing is going to command interrupt me every time an alert comes through that I have to pay attention to and regain the context of whatever that multiple projects stream, you know, has risen to attention. So you're not at the computer, but the fixing the, the 50 fence posts took three times longer because you kept stopping to tell your agents to keep moving. And you're like, but look at me out in the wild. And we're going to be like, yeah, but it was actually more efficient for you to do five hours at the computer and then be gone. So this is something that, uh, comes up in improv, uh, and specifically musical improv. We work with musicians who play, who are the accompanists and we're going to make up the, we're going to make up the songs on the spot. Right. And the musicians would say, Hey, when you're off stage, be thinking about the melody that we sang and what harmonies you can add to that. The performer off stage is watching what's happening on stage. Cause when they come on, they were in this world, they know everything about this. And so it's, it's an assumption that Andy is walking around on his ranch that he has the attention to pay to the AI when actually Andy had a full attention for things that he was supposed to be doing on the ranch. Right. I mean, you are literally walking around looking for where the horses did something, or is there a problem with this over here? Right. There's a, I don't know what you're looking at is the root coming out further. So that's going to be a problem, right? Those are plenty of deferred maintenance. That's right. Right. There's, but those are tasks that are already in the box. It's not like the box is empty and you're like, sure, I'll just like have a conversation with AI at the same time. It does mean that that's crowding out other things that were perhaps also life affirming. And yeah. Yeah. This is a, we have neuroscience tells us that we do not have multitasking availability. We have single threaded attention. And anytime an interruption comes along, you drop what you, you know, you were, you were processing and you lose that to regain it has an overhead. Yeah. And so if you end up being sort of chaotically factionalized into all of these different demand spaces, that's not an efficient way. And that's why coders speak about a flow state that should not be interrupted. You have to maintain that concentration on a single context, not multitasking. What were we going to say, Gareth? I liked your comment. My ADHD says otherwise. Yes. I'm great at multitasking. Wait, wait. Multitasking at the same time? Yes. But yeah. And then what. Like this hand's doing one thing while the other one's doing another. Oh yeah. I do that in the kitchen all the time. I'll be shutting covers with my foot and putting away dishes with my hand. Yeah. That's a normal thing for me. I think something that does or Dot does a really good job of is like just kind of sitting down in the background and not saying anything unless spoken to. So I could and even I can talk to my kids and Dot won't respond. But then when I talk to Dot, it will respond. And so I think that's and so I can call Dot. I just call Dot and picks up after 10 rings. I don't know why it takes Dot so long to pick up like Dot's busy or something. Um, like sometimes he doesn't pick up. 10 rings is a lot. I know. I'm like what's taking it so long to pick up? But it's I don't understand it. But it's whatever. I mean, it's more realistic. Right. Um, so but I can talk to my kids and then all of a sudden I can talk to Dot and Dot will talk to me, um, randomly throughout the day. Um, and so but they kind of goes to the multitasking. You could go on a walk and not say anything and not have to talk anything. And then all of a sudden Dot will come in and say, yo, this is what's going on. Okay, cool. And that's like a perfect assistant. I think they do a good job with that. Um, yeah, that's I don't I already have the comic for this week's newsletter, but that is a funny comic idea that Dot is like just not picking up, you know, an old school telephone with a with a cord. And it's like another the other dots like you're going to get that. You're like, that is a funny. That is. Yeah, yeah, yeah. He'll be all right. He's going to stay there for all 10. Watch this. That is a great idea. I might have to make that today. Yeah. Let's talk agents for a minute here. I want to talk about new research like that's N-O-U-S Hermes agent. That startup just raised 90 million dollars at a mere 1.5 billion dollar valuation. Now, why do I say near? Because new says it's open source Hermes agent, which you don't pay for. I gather. But still, you know, people give much higher valuations for something that actually has enough traction in the marketplace to offer opportunities for monetization. It accounts for roughly two and a half percent of global token usage. So Hermes as an agent is intermediating agentic work that is driving two and a half percent of total token usage for all computing in the world. That's pretty amazing. Now, that's an internal assertion from new research. But who knows? Like it clearly is, you know, gaining some traction out there. So it's gotten a lot of traction with software developers and with individual users, Bessie user. I started mounting it on my machine last night and I was surprised to find that while it provides a long, long list of the various models you can set it up for APIs with, it doesn't include Anthropic. So you can't set it up with Anthropic. And I was like, so I spent hours with Claude yesterday evening trying to get my Claude API connection hooked up to Hermes and finally gave up. I had to go to bed. It wasn't going to work. And so I said, OK, well, I'll use an open AI API key tomorrow and go there. But I don't want to do that because Hermes on their Hermes index now shows which models actually perform best in agentic tasks using the Hermes harness. And Claude is at the top. OK, so somebody out there has figured out how to use Claude with Hermes. And so I want to use Claude with Hermes as well. Anyway, I'm working on it, but I'm not having a very easy setup for Claude plus Hermes. So in the history, everyone wanted to use Claude with Hermes, right? It was great. And you could use your account the same way that you could set up your open AI account to use it. You just go in and says, yes, we can. We authorize the ability to use the subscription through this. There are other systems that do that, too. Anthropic sort of famously was like, no, we are so good that everybody is doing this. And Hermes is really efficient, right? Like you can have Hermes proactive. So even if it wasn't called token maxing, you were token maxing because you gave Hermes a bunch of tasks and it continued to do that. Anthropic was like, nah, that's not going to happen. And they came out with something that says you can now hook up Claude. I don't think it's with the subscription. I think it's part of the SDK. But I agree that that's part of the rumor that you can do this. But it is obviously difficult because there are a lot of things that I see in Reddit. Like, wait, I thought you could do this, but how? Someone who's making it work, tell us. Which is actually relevant to Claude. Anthropic just announcing that they're giving a monthly credit allotment to the max subscriptions. So the 5x max, which is not 5x of anything, it's just its name now, is you get $100 in, I think, API credits in addition to your subscription. And 20x gets $200. Starting to say what 20x actually is. Maths, sure. Sure. Let's do two other quick rapid fire Anthropic. Because Haiku 5.5 came out and they say it's 75% cheaper than the last Haiku. And up to 90% cheaper on shorter requests. It also halved the price of Sonnets. The other secondary news to that, so that was Haiku. And I think there's a lot of good areas where that could work. That's $0.10 per million input tokens. So that's what they have there. But the other part that has actually already affected some of my work on Amanda's Ajova Journeys project is Sonnet's 5.5 cash reads, which they knock down by 50%. Cash will get you, man, through the API. So I have already seen a reduction in costs on some reoccurring agents that I have working in the background for Ajova Journeys. Basically, every three days, Ajova Journeys, the command center for the business, goes out and searches for new ideas for Amanda to make videos. And it's a very lengthy list of... It's reading all the YouTube comments, all the TikTok comments, what other cruise people have made content about news. The whole thing, right? It goes and looks. Well, a lot of that does end up being part of a cash problem. And so you end up paying for that too. And it's not huge, but I've seen it go from about like $0.85 to do this initial idea burst down to like $0.75, $0.70 in the last couple of ones I've done. And from what I can at least see, that would be from cash because nothing else has changed on there. So that's over hundreds of runs that starts to really make a difference. So I just wanted to bring that up as well. So IQ 5.5 is out and Sonnet 5.5's cash reads are now 50% less. And then also what you just said, Beth, as far as the monthly API credits. So a bunch of different stuff on there. I do have one other quick thing. Just as more of a promo, but I'm going to tell people they should definitely go check out this Conundrum episode coming out on Saturday. The reason for it is, is something I think everybody can relate to. Now, I tried to do that with every Conundrum episode, but it is about AI agents. We've touched on that before, but it's more about what happens when everybody has AI agents. So it uses the example of let's say concert tickets, but it could be reserving an RV site or a very popular restaurant, pretty much anything where supply and demand are, there's much more demand than supply, whatever that might be in life, right? So it's an interesting Conundrum because we can see a future where having personal agents to go get, I'm going to use Olivia Rodrigo because I just went through this for my daughter. That is the hottest ticket out there as far as my world is concerned. That's my daughter and all her friends. They just want to go to Olivia Rodrigo. Maybe not as big as Taylor Swift was, but certainly the biggest thing going on right now for her. By the way, I love Olivia Rodrigo's music. It's awesome. So that's what we were dealing with as a family and trying to see if that could work. Okay. So take a future state where her tickets go on sale, but everybody's got agents. Okay. Well, I remember being a kid and asking my mom to take time off of work because I was going to be in school to go stand in a line over a Turtles music store to try to get us tickets to some concert, blah, blah, blah. Right? So that was always a thing. And nowadays it's all online. Same deal though. Tickets go on sale usually at 10 AM, 11 AM. Lots of people are working. I've had in-laws that have tried to do it for my daughter and other times where Amanda and I just were not available. So like, this is a thing, right? So you go, okay, one side of this conundrum as you listen to it, I definitely recommend people going and checking this out on Saturday when it drops at 10 AM Eastern is because on the one side you go, Hey, great. I don't have to be available. My agent can go be available on my behalf. That levels the playing field for all the people. And we're not talking about bots here. We're talking about everybody has an agent. So it's on the up and up. Andy has one agent. I have one agent. It's, it's working on our behalf. So on and so forth. Okay. So you go, okay, that's great. But then you go, okay, well, hold on a second. The other problem, the other side of that says, the conundrum says, well, if we all have agents, then it costs me nothing more. If I give my agent a preset budget and let's take a RV situation. Well, what's the stop my agent for giving it a shot and seeing if he can get a ticket or an RV spot or that restaurant. So now instead of me, me going and saying, okay, I have time at 10 AM to put my attention on this Olivia Rodrigo to see if I can help my daughter. She's in school, whatever. Instead of that. Now I have an agent that's like, I'll happily give a shot at anything. Even for stuff you don't really actually care about because Hey, what's the downside of that? So now you go, well, okay, damn. It's evil even for everybody, but also it means that I don't have to be around for any of it and concurrent things going on. My agent can handle that on my behalf. So you can see where this is going. Now that doesn't even touch what's going on right now with the ticketing agency and bots and latency, but latency becomes a big problem as part of this conversation because milliseconds, just as it is today with the bots start to become a bigger deal. If Gareth lives closer to a connection point, a data center or whatever, and his, his agent for speed alone, for latency alone can simply grab those Olivia Rodrigo tickets. 400 milliseconds faster than my agent can connect and do it. He still has a competitive advantage, even though we both only had one agent in the ring, right? Have an option. So then it brings up the future. As you go through the episode, what I would love people to go listen to is more on the, on the backend too. It's about a 20 minute episode, but what's great about it is this are saying, well, what do you, what do you do? So is it lottery systems? Is it lottery systems? But also, cause again, my agent can apply to 400 lotteries. There's no harm in that, right? Hey, go give it a shot. Hey babe, we just got a, we just got an RV site. Do you want to go camping? Cause we have it. Is it deposits into a library, into a lottery so that there's money on the table and you have to spend it if you win the lottery. So it gets really, really interesting. I think we already solved this problem because of, but. We didn't solve it cause there's a conundrum about it on Saturday. Maybe after Saturday. You can try, Garrett. I'm kidding. I'm kidding. Go ahead. I just cut back to earlier the thing. Garrett said he doesn't manage people cause he's a little too blunt. I think we're seeing it. Karen says you don't need the conundrum episode. No, go ahead, Garrett. No. Because Nike, Nike had to do like a big thing to stop resellers and bots from getting shoes when they get dropped. Right. Way of blocking people. And then anybody who does want to purchase a shoe also goes into a lottery. And to get the shoe, I never win the lottery. Don't know why. I've tried many times. And so I think that's, I think probably more companies will have to adopt that. But I did think about this actually this morning when I was like thinking about Nike and I was like, well, can I just have, cause I wanted some shoes. Can I just have my, my agent just log in and buy the shoes for me at nine o'clock in the morning automatically. And it can pass all the security things. It can bypass everything that it needs to, cause it's using my computer. It's not using just a system. And so, um, yeah, it's, but you only get one entry entry and you, I mean, they know if you're trying to go and I mean, they use your IP address, I'm assuming to do it. Um, and so, and your email and stuff like that. And so, um, I don't know if it could be solved, but it is that competitive advantage because not everybody's going to have access to these agents. So that'd be a first problem. Correct. Yeah, I agree with you. That's the first problem is anybody with agents in, you know, that's in a, uh, you know, rich society, right. That can afford to do it. Does it mean that because I can afford a hundred dollar clog code that I have a competitive advantage over everybody that does it? Well, yes. Right. But it, but what I'm saying is in some future state, most people will have agents at their, at their ability to do something on their behalf, be it a muse, be it a whatever. And so how do you solve for this? So anyway, I just found it to be fascinating. I mean, please go listen to the episode. Um, I often say this as my intros of the show is like, this is one of my favorite episodes. I really do think this is my favorite one because I think it's so relatable to so many people. And there are just so many people who would love to see Ticketmaster burn to the ground. That's not going to solve anything, but there's lots of people who just want to see them a dearly. They see them as the problem. And a lot of people have felt this in some way, whether it's a restaurant that you really wanted or shoes or RV sites or camping or, or Olivia Rodrigo. Right. So it's something we can all really, really relate to. We've all been on the outside looking in. And so how do agents play into that? What does it solve for us? What, what extra problems does it cause? This to me is why the conundrum episodes of my, I love doing them because it's just like, Hey, this is coming. What do we think about this? How does AI impact this for better and for worse? And are you okay with it all the way up into the point that everybody else has what you have? Meaning, are you okay with it? And you're willing to turn an eye because you have a faster computer because you can afford that a hundred dollar plan. Is it okay that I can deploy my agents and get the shoes and Gareth, you can't, or should I feel bad about that? That now I have a real competitive advantage, but it's just like, well, it is what it is. I guess Gareth's on the outside looking in and I got the better deal here because I can use my AI to get around. Speaking of concerts, if you're like a fan of seventies things there's like seventies classic rock. And then there's also 70 plus year old people who are going to gather around rush. And this coming week, this coming week, Jude and I are going to the rush concert in San Jose and you know, rush is doing a, a major us tour right now. Uh, and, uh, we got tickets close to nine months ago. I think when they first launched that thing. So rush and the day before we're going to go see Bonnie rate just to kind of put all of that into the context. And those people, it's, it's amazing. These are old people like me who are still performing at, you know, uh, sort of high levels and you're doing tours like that. Bonnie rate and brush. Uh, that's a lot that, you know, physically and cognitively demanding to do those performances night after night, or maybe about tonight. That's sort of two months ago. I went and saw John, John Mellencamp cause friends of ours had gotten free tickets. One of the best concerts I've seen probably in the last few years. And I think he's 67, 68, maybe 70, somewhere in that range. Um, and he was phenomenal, phenomenal the whole time. So yeah. Uh, Tom, uh, Tom Sawyer is by rush, right? Yes. That's the big, that's the famous rush. That's, that's a very famous song by rush. Checking my brain there. Uh, yeah, very good. Rush was, I was a little after rush. Hey, I've got a couple of little things that are. One is that chat GPT has been making cartoons in the style of the New Yorker, uh, artists and signing them with the names of the real cartoonists who never drew them. Yeah. And this was an investigation done by the Niemann lab and published on Monday this week. And, uh, a reporter, Andrew Deck found the signatures or pen names of more than 15 contributors on AI generated images. Meaning their, their art was, was captured by the AI and attributed. Not, not really, but it was, it was basically the AI was generating it as if that was authorized by the artist. That's another little twist on the whole question of whether there's, yeah, go ahead, Beth. No, no, no. Because, um, I, it's a, it's a twist on the, on the piece, but it makes sense because for the training, the signatures, part of the image, all of their art, right? Includes the signature. And, um, back in the good old battle days of Dali, um, that was happening to images. I was generating a bunch and by was doing all kinds of trippy styles, mixed, you know, doll, uh, actual Dali, Salvador Dali with, I don't know, uh, Mary Cassatt. Um, and then like, yeah, let's see what that art is like. It would often generate some sort of smear of things that look a little like letters, because again, that painting should have some name there. Yeah. Good. And this other one is, is interesting. And I think you guys might've heard about this, but a small studio called Artcraft used Claude to make copies of all of the Adobe software platforms. Now they, they've only released, uh, two of them in kind of full form and their photo craft is a Photoshop clone, right? And it's the furthest along. It supports layers, masks, adjustment layers. And the developers say that of 135 real world PSD test files that were loaded by the new Photoshop clone, that they came back in, in perfect form, right? So the, the, the system is, is working as if it is Photoshop. And they say that this was a quote, quote, clean room build, meaning they didn't use any code from Photoshop. They just said, okay, Claude code, build me something that does what Photoshop does. And they, and it's working. And they built it in rust, which in terms of code, you can, you can see people saying, you know what? Fine. This is really messy code. We're going to do the thing. We're going to build it in rust. It's going to be really more efficient. So that's like a button to this kind of stuff. Kind of funny. Yeah. So they have all of the, of the tools coming out, but, but photo, so Photoshop illustrator, premier light room, after effects in design and Acrobat pro. Well, I, I want to get the open source clone of Acrobat pro because I'm so annoyed all the time when I try to open the PDF and I have to go, no, open it with the Chrome extension, you know, in order to editing a PDF. Yeah. So instead of having to pay Adobe to have a PDF, which is kind of a stamp, I think wasn't PDF originally done by Adobe. I mean, it developed that standard, but now it's an open, go ahead. Yeah. They had to, they had to release it. Right. That was a, that was a monopoly or problem. Yeah. Yeah. Problem. And so they had to release it to be able to be worked with by other things. And I think maybe it was because that was happening in a browser. Right. There were, there was, it was a while ago. Rush was really popular at the time. And that's funny. But all of those are GitHub repos. Like you can go to whatever the art craft or whatever their, their business name is. And art craft is the name of the studio. Yeah. You can see what they've done and how far along they are and where, and where those files are. Yeah. Exciting. Greg came in. He goes, what did I miss? I said, the show, you missed the show, Greg, but he goes, rush. Also, I want to call this out. Gareth had put in the comments, something on my desk just moved by itself. Then Gareth left. Yeah. We don't have any backroom chat for why Gareth left. So you should definitely, definitely tune in tomorrow to find out what happened to Gareth. Because what moved on his desk and why did we lose Gareth? On the live stream. You won't know until tomorrow till we can see. And we can tell us that. I, yeah, I don't know. I just, but yeah. Anyway. Anything, anything else? I mean, it would just a minute or two here, but anything else from you guys? Otherwise we can wrap. No, I'm good. So the drop of Haiku 5.5, right? Anthropic dropped Haiku 5.5 yesterday. I'm seeing people excited about it. It seems really good. Um, but what I'm watching for, and I'm not sure it's going to make it is, um, is so for open AI, you can use Luna max and kind of get at the level for less than you would, um, for one of the higher end models. So I'm watching for the discussion about, can you use Haiku on a high effort level? And what does that get you? Um, because, because Luna is like good. It's fine. But anthropic models are just better for some kinds of work. And that would be interesting for me. Like it's, it's still delightful. I am absolutely using Opus 5.5 on medium and doing a ton of stuff. You talked about overnight things. I now have a honeydew list, right? Like, okay, that's the honeydew overnight. Thank you very much. Um, so that we set it up before and was like, okay, just run it, uh, honeydew LFG for, uh, compound engineering. Just like, okay, show me in the morning what you have finished. Thank you. Um, but that, that kind of, that kind of thing to be able to, um, spin out some sub agents with, you know, a narrow super intelligence, um, of Haiku, that could be really exciting and impactful. So, yeah, I agree. So there was some stat about Haiku being somebody quoted it. Like it was super helpful with like, uh, basic CRM tasks. And I'm like, oh yeah. Okay. From, from like the scaled side of the house, um, Haiku plus like a, a zoom info or Apollo or a clay could be really, not clay. Clay would be expensive. Zoom info probably wouldn't be because you're just pulling the API, but I'll tell you one of the problems, like so many of my clients have is just like people move, people move jobs and your CRM data gets old. And if there's a cost effective way to have like an agent that just goes in pings and goes, all right, zoom info, what do you have? Oh, it's different than what we have updated. That's a great one by one job for something like Haiku. What's not a great job for Haiku never has been is like right. Give me a whole dissertation or even like a, a three page paper, right? To me, that's where the context window would just, it would just go to like all sorts of hallucinations and stuff like that. So I do think like, yeah, it's kind of like the Jev use case, but different because it's still an LLM and it can still hallucinate, but it could be really good for things like that. So yeah, like I never had the human ability to go check all these, but I would love it if my CRM was up to date. Oh, but I pay for this other subscription of, of like a, you know, fact checker, like a zoom info or an Apollo. Okay, great Haiku. And I would test that over like two, three days and check my costs. Right. Why not? Speaking of, I mentioned earlier on the show that I was, uh, you know, flummoxed and blocked from using, uh, anthropic models with Hermes agent. But on Hermes own index, they show number one is, uh, is opus 5.5. That's the one that, you know, does the best job with Hermes, like in agentic tasks. And then there was somebody else in between. And then sonnet 5.5 was the third in the list. And it only costs like, you know, I forget what the metric was, but it was like a very much, it was less than half of what opus 5.5 would cost using the Hermes harness. So I thought, man, I want to use sonnet 5.5 because it's cheaper with my Hermes agent. But now I'm hoping that maybe there's an even cheaper alternative out there that has come along. And just in the last couple of days, we talked about Mistral's LeChonc. I think we talked about this yesterday, which is really very, very akin to the frontier models. It's a trillion parameter model and very inexpensive. So hopefully Hermes and other agents can access LeChonc and do it that way. But I'm not sure how to, uh, how to evaluate that because Hermes does provide a very long list of alternatives, but it's, it's missing some. Yeah. It takes a, it takes a little while for models to catch up. You can also run a local model. And I believe that's what Cisco is doing. He's, uh, he's running 3.8 flash. No. Hi Cisco. We know you're watching afterwards. Miss you. Um, yeah. Um, we're good. Good. Yeah. Good. Good. All right. We'll call that for our Thursday show then. Um, thanks for everybody for hanging out with us. Thanks for everybody in the comments as always. I don't forget, uh, to like, and subscribe. If you like what we're doing, check us out in the Slack community at the daily ishow.com. Uh, come join us live. If you're available on YouTube during the 10 AM Eastern, uh, live time that we're on the show. And, uh, just so that we do this five days a week in case you're just catching us a little bit later in the show and you're, you're popping in. So, uh, we're here, we're here five days a week and come, come hang out with us to, to discuss all the AI news. We've been doing this well over three years now. Uh, 829, nine, nine. No. Yeah. Nine. Is that right? 829th episode today. So there you go. Come back for 830 tomorrow and for all the other ones. Uh, until then guys have a great day. Thanks to Garrett. He was here. Bye.