# EP #43 Is the Cost Apocalypse Already Over?

“EP #43 Is the Cost Apocalypse Already Over?.” AI at Work by AscendAI, 2026-09-24. https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/

Credit the named speaker when quoting a passage, and link to the episode or the passage.

Transcript licensed under [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/).

### [Kevin Williams · Passage 1](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-1)

Hey Matt. How's it going?

### [Matt Graham · Passage 2](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-2)

Hey, what's up, Kevin? How you doing, man?

### [Kevin Williams · Passage 3](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-3)

I I'm doing well. So you know, a lot of people listen to us and they can't see us, but we managed to secure our first podcast sponsorship. So I mean, you know, that's like the big time, right?

### [Matt Graham · Passage 4](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-4)

Dude, w this is the big leagues, Kevin. We made it, man. We got sponsors. Woohoo!

### [Kevin Williams · Passage 5](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-5)

So so those of you paying attention on online, you'll see this fancy chair behind me, which is a a Liber Novo chair. And I have to say, this is the most comfortable chair that I think I have ever sat in. And that's not just because they're a sponsor. But I have mixed emotions about this.

### [Matt Graham · Passage 6](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-6)

Tell me more. So I tell me more.

### [Kevin Williams · Passage 7](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-7)

So the idea of this chair, and the I promise, guys, we're not gonna be like oozy about sponsors or anything, but it's like a super supportive coder chair, super duper awesome, right? And it has a like a fan that's built into it, it has a back massager. But the reason I'm torn about it is because I apparently am spending so much time in front of my computer, like vibe coding and building things, that the world noticed and they needed to give me a highly supportive chair.

### [Matt Graham · Passage 8](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-8)

Yes. this chair is means we can spend so many more hours at the desk coding with our favorite Astra or Cl Fable. Like man,

### [Kevin Williams · Passage 9](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-9)

Yeah.

### [Matt Graham · Passage 10](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-10)

just imagine being able to lean back and just like code for another ten hours 'cause you're just so comfortable in this thing.

### [Kevin Williams · Passage 11](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-11)

I I I do feel a little bit like some of the characters out of WALL-E a little bit. Like at some point, I mean nobody can see this, but I have like five monitors around me and now I can like kick back in my chair. And since I my MO is to spend so much time just talking to Claude or to Kodak's anyway, I don't know. I think I it's it's awfully close to Jarvis.

### [Matt Graham · Passage 12](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-12)

Yeah, this is the thing. Like you can lean back and like typically you can't type very well, right? Like when you're like in that kind of position. But now like the like with whisper flow and all these the voice commands are like so good. The thing I like, okay, so if you're not down with the chair and leaning back and getting like all wally on everything, go for a jog. Like you can go into voice mode, connect your, you know, I'm sure a lot of our audience have done this, connect their their chat GPT on their phone or their cloud on their phone to their remote desktop in. Essentially that's what it's it's kind of like. And you can just call on every function you need, every tool, MCP that you set up or full on computer use from your phone while you're jogging. like, you know, so if you're not feeling the chair, Kevin, you can go out there and jog and still get just as much work done.

### [Kevin Williams · Passage 13](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-13)

I love it. And it's a beautiful day outside. So I'm gonna have to do that. But actually elaborate for just a second. So I'll admit that that I, of course, use remote through GPT and through Claude, and it helps me track my ongoing processes. I totally shouldn't be doing this when we're doing a podcast, but I I have two processes going on Codecs and probably three going on Claude Code right now. And if I wanted to, I could open my phone and I could sort of shepherd those through and get a sense of what's going on. They need course correction, they need approvals, whatever. But I don't think I'm actually quite getting how you can talk to it during a jog. Like what what does that actually look like from an experience perspective?

### [Matt Graham · Passage 14](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-14)

Yeah. So that it's a little facetious because when you're talking, like, you know, you're in voice mode and that's really gonna be better for chat, not for like running codecs or, you know, running chat GPT work, right? But you can just you can use the the voice to text, right? And let's say I was at the conference last week. Okay, I was at Unbound in Boston, it's a big HubSpot conference, and you know, I there's all these like 20 minute tasks that I need to get done. That I can I just usually wait until I'm back at a desk for. but I can just remote into my Chat GPT at home. I left my desktop running 24-7, doesn't go into sleep mode. And let's say I met somebody at the conference. I immediately thought of two people that this guy should connect with, but I couldn't remember who they were. So I remote back into my desktop at home through the ChatGPT app. I ask, hey, I had a call with these two people around these particular dates. Go into my call recording system, go into my calendar, look around and if you can see if you can find their name and their email address. Then once you do that, draft some two introductions for this guy that I just met. You know, I don't have his LinkedIn handy, but this is his name. This is what he does. Go look him up, you'll see. and you know, that all that like clicking around to try to remember who those two intros were, like what was their email address, like what's the angle here for the intro? Cause these are, you know, you don't want to be wasting people's time, you want to make a thoughtful introduction. All that would have taken me like 10, 15 minutes to go search for all that. And I'm at a conference. So like my to-do list is adding up. So, you know, then I come home from the conference. A bunch of stuff is already done because I could just talk to Chat GPT quickly in a pretty garbled way with lots of background noise and pretty loose instructions. And it could go take over my computer to do all those things. Now, some of those things it can do through the MCP or through a plug-in and a connector, but some of those systems like the call recorder. It had to go into two different call recorders and one is plugged in, but one's on the browser and so that's why it has to take control.

### [Kevin Williams · Passage 15](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-15)

Okay. Okay. So I so I buy that. I mean that's that that that I think is this next like form factor that we're going through. But the audio one is fascinating to me. And when you hear about like Johnny Ive partnering with OpenAI, Johnny Ive being the the the famous iPhone designer and all the billions of dollars that open AI paid him to come on board, I'm pretty convinced that the form factor of it is going to look a heck of a lot like the AirPods that I see Matt wearing right now, such that you're not even going to open your phone. You're just going to have some sort of a trigger phase, which I won't say right now if you say the S word or the A word and all kinds of devices freak out. and we get some hate mail about that. So yeah, but it will be something along those lines. So you're at the conference, you're at Unbound, and you just say, Hey GPT. X, Y, and Z, and it's gonna go off and do the thing and it'll probably whisper back at you, okay, that's all lined up, Matt, and then moving on. And meanwhile, Matt, the this coding project that you have has run into an objection. this is what the objection looks like. is that something you want me to handle, or do you want to save that for the next time you're back at your computer?

### [Matt Graham · Passage 16](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-16)

Exactly. So you know, voice mode is like really for just like quick conversations of like, hey, teach me about this interesting topic and you're usually in chat mode for doing work, whether it's codecs

### [Kevin Williams · Passage 17](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-17)

Mm-hmm.

### [Matt Graham · Passage 18](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-18)

or in actual work mode, like you're not you're it's not gonna really talk back to you very easily. Usually it pumps out a bunch of stuff and you have to like sit there and read that. But it will be interesting if they can make that shorter and s more succinct in a voice mode model. but in the meantime, like if you just have quick thoughts while you're on a jog or on the go, dump it in there. And then see, you know, what it comes up with and you can glance at your phone and get it unstuck every few minutes or, you know, every thirty minutes, whatever.

### [Kevin Williams · Passage 19](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-19)

So I think this lends itself to something that I think is a is a pretty main topic for us right now. And it's the it's the proliferation of different access points into LLMs and how consumer facing a lot of them are are rapidly becoming. the one that's getting the most news right now is Meta's Muse product, which is It's essentially an embedded agent right in the meta universe. And it displaced OpenAI as number one in the app store almost immediately. So it's getting millions and millions of downloads, which to me suggests that it's stru it's struck a certain consumer chord with people. And I think diving into this like universe of semi-autonomous agents and What's connected and what isn't connected could be something pretty interesting for us to explore. And I actually spent several hours this week playing with Muse. So I have some some pretty strong I guess impressions slash opinions so far.

### [Matt Graham · Passage 20](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-20)

Well, I'm super excited to hear that. But I think it goes back to our conversation last week that agentic AI feels like it made like a really big step change, especially personally for me with Astro, its ability to use a computer and unbl like, you know, with OpenClaw and the prior harnesses that were available, they'd get stuck. They would end up going rogue, they didn't have a lot of guardrails, or they would just get stuck every five minutes and you're just spending more time getting it unstuck versus just doing the task yourself. So I I personally feel like this month. We saw like Muse, we saw GrockBot, we saw

### [Kevin Williams · Passage 21](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-21)

Mm-hmm.

### [Matt Graham · Passage 22](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-22)

Astra. And like it feels like a big step change, especially for people that, you know, they can't pour tons of hours into this stuff. Personally,

### [Kevin Williams · Passage 23](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-23)

Mm-hmm.

### [Matt Graham · Passage 24](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-24)

I think the onboarding and the setup is becoming much easier, much more approachable. And so we're seeing it sort of cross the chasm. It's definitely not across the chasm yet, but I think we just made a little leap into another, another area on that curve where there's more mass adoption. So I'm I'm super pumped to hear your experience with Muse.

### [Kevin Williams · Passage 25](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-25)

So so so let's back up for just a second, and it's like a brief history of agents in 2026. so

### [Matt Graham · Passage 26](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-26)

Yeah.

### [Kevin Williams · Passage 27](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-27)

everybody disappeared in November of 2025, went away for the holidays, and they discovered all of these new features that had been baked into Claude in particular, as far as new powers. And so they had a little bit of free time, geeks being geeks, they spent their holiday like playing with X, Y, and Z tool. And then in January and February, you saw a pivot to the first real autonomous agents. And that was those were the days of OpenClaw. And OpenClaw very quickly took over the internet, not literally, but burned billions and billions of tokens. And it was it was the first like agent that was more configurable. Outside of the sandboxes that the Agentic Harnesses and the Frontier Labs had really put together. You could connect it to a whole bunch of different data sources. It had a heartbeat to it that would like loop back and make sure that things are going well. Like a very simplistic version or example would be like an email heartbeat. So I'm I'm not just going to look at your email box once and triage it. I'm going to look at your bot email box on a 10-minute increment or a 30-minute increment, and I'm going to prioritize things and I'm going to bring them to your attention in a form factor that you're already using. So you could pick and choose whether you were using Telegram, which yes, Telegram does still exist. this probably gave it a a lease on life. Or if you wanted to use Discord, or if you wanted to use Slack, like you could surface these heartbeats in ways that were much more proactive. And that really opened everybody's eyes into sort of the art of the possible. that got really problematic in a hurry. everybody had to go like run out and buy Mac mini minis. And like apparently Apple made a hundred million dollars on Mac minis. Thank you, OpenClaw. and then That just idea started percolating. So the next piece you had was was Hermes. And Hermes sprung up as being a little bit more controllable, had better model selection, et cetera. But both of these were still deep, deep, deep in the realm of the geek and the ability to like configure them. And I think that there was just so much hype around it, like, okay, we're AI influencers too, right? But it felt like All of these people were like, you know, Hermes and OpenClar. I'm running like 80 different agents and they're doing X, Y, and Z. And I always had sort of a show me the money like aspect to it, like, all right, okay, sure. You have this desk and it's marked marketing and it's a mi op Mac mini and it's supposed to be your marketing desk. But the actual outputs were pretty kind of sloppy and whatever.

### [Matt Graham · Passage 28](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-28)

Yeah, I I I would say it's sort of like, you know, when self driving came out ten years ago. Like, sure, it's really easy to have a self driving car run around go in a loop around a circle or track, right? And everyone's like, Wow. You know, that was like our moment in what, like February of twenty twenty six when we saw Open Claw and what it could do. But like driving in a loop on a racetrack, self drive it's not self driving, right? Being able to drive through Boston's like two hundred year old roads or whatever, you know, like I it's a totally different story. But keep going, Kevin.

### [Kevin Williams · Passage 29](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-29)

But the idea was sound and the idea of being able to have agents of agents and the idea that you would have an orchestration agent that would be able to talk to the other agents, like all of these are what you would call a primitive for like the technology that's developed since. And then the next one, and there's a lot of in-between next ones in there, but the next one that really matters that came to the table was Grokbot. And that was really just a few weeks ago. And boy, this is like time just flies with this stuff. But

### [Matt Graham · Passage 30](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-30)

Yeah.

### [Kevin Williams · Passage 31](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-31)

it took a lot of the different lessons of of how Hermes and OpenClaw worked. And it just made it so much simpler to be able to sign into GrockBot, to be able to sign into different services, to have a common interface and be able to start actually doing functional things, like I need to update Search Console as a marketer. So as opposed to flailing around with Search Console, I'm just going to Point Grokbot at it and tell it what I want, it's going to basically use computer controls in order to do that. And then it's going to report back. And then you might have another Grokbot agent that's like a marketing orchestration agent that is looking at your search console and what you've done there. And then it's incorporating it into the whole. And then maybe once a week that bot comes back and reports to the marketer, hey, this is what's going on in your marketing environment. What would you like to do next? okay, Grokbot. I see that our search efficacy is slipping on these terms, which by the way, you're doing kicked back in your cool chair. We're just like talking to Grokbot. but you're you just say, hey, Grokbot, we need to build you know, a content strategy around this. And I'd like you to sketch out what a blog is gonna look like and et cetera. Make it so, and off goes the bot. To go do the thing. And I think it's kind of important to add a like a punctuation point here, because there's not a thing I just described that somebody with some decent skills in Codecs or Claude Code couldn't build using edge functions and some of the stuff that we've been talking about here for over a year. Automations, being able to do things on what's called a cron or a schedule, and having the benefit of like a central database where all of this stuff goes that you can tap into. But the bet is a lot of people want to move a lot faster than that structure will will allow. And I'm not, I'm sort of skirt on on on the middle of the line there as far as which is the better way to go. But it's certainly interesting to see the way that this is developed.

### [Matt Graham · Passage 32](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-32)

Well, I think it's like the it sort of goes back to that lovable argument we had a few weeks ago about approachability. I think that's a always

### [Kevin Williams · Passage 33](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-33)

Yeah.

### [Matt Graham · Passage 34](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-34)

an underweighted aspect of these arguments of which one's better. Well, of course there's different pros and cons and feature sets of all these different tools, but really I think people aren't considering the ease of use, you know. Some things are just much, much harder to use if you know, like you know, people still get intimidated by clawed code and codecs. So I I think Grokbot and maybe I haven't used Muse yet. So I'd be curious to hear where on your history here Muse is. But I think Grockbot is just a lot of people that aren't that technical are pinging me like, man, I'm so excited by Grokbot. And it's just because of the ease of use to set it up and make things happen.

### [Kevin Williams · Passage 35](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-35)

Well, yeah, I think most normal people want things to be easy. This is this

### [Matt Graham · Passage 36](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-36)

Okay.

### [Kevin Williams · Passage 37](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-37)

is the way that, you know, Apple has become one of the biggest companies in the world is by making everything easy. another sort of side battle that's going on in one of my communities is wow, Android's gonna take over the world because it's gonna have small models, it's gonna have Linux installed, it's gonna be able to do all of these really cool things. And my response to that is meh. Apple's still gonna be dominant because it's a luxury brand that makes life easy for people. And yes, they kind of suck in AI, but they're figure it out eventually. And yeah, that's terrible. It's absolutely terrible. but but most people, most normal people, like want to be able to have an experience that's really clean and easy. And this is sort of where Muse enters the scene because Muse is a little bit of a different beast. even though XAI, which is grok, is Predicated on consumer communications and technology through X, right? GrockBot is actually a business tool. And it is positioned as a business tool to be able to accomplish business tasks. And what that left open was the ability for Meta to come up with Muse that is a that is a very accessible consumer solution to this same challenge. And as a business user, I think it's important that you keep that in mind. When you're assessing the different platforms. Who is the user? What is the user case? What is this thing actually supposed to do? and calibrate your expectations accordingly. And so Muse is the same idea. Like, let's connect a bunch of our data sources, let's connect our email, let's connect our credit card. Like, yeah, really? Like people are are signing up with like their MX login and like having it like scan for things. And

### [Matt Graham · Passage 38](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-38)

Let's go. Hundred K limit.

### [Kevin Williams · Passage 39](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-39)

And basically it's your little little personal assistant that you would think that an initially it would be very calibrated to Facebook, and it is to a degree like, hey, help me find X, Y, and Z in Marketplace. This is this is what I'm looking for very specifically. Set up a search that runs once a day and then pings me when you find whatever the thing is. Okay, well, that's that's not very exciting. But how about it this is straight from the Muse like suggested activities section. how about log into my Amex account, look for my spending patterns and try and figure out what Amex rewards I should be enrolled in or scan my email and and figure out what subscriptions that I'm probably overpaying for and then help me triage those I want to unsubscribe from. Okay, these are still kind of pedestrian things, except Muse will actually go the next step. it can be enabled. And by the way, this is not out of the box, but it can be enabled with phone capability. So what people are doing is they're saying, my gosh, my my ATT bill is really high. I'd really like to figure out how to lower my ATT bill. So they'll have Moo Muse go out and do research, figure out like how their account can be reconfigured. Apparently, ATT has a As a rebate for direct pay. So you save $10 a month align, but they don't really tell you that. So Muse will call customer service on your behalf and suss out the details. Now, whether customer service is actually going to talk to the Muse bot, which by the way, doesn't pretend to be you. It says I am Kevin's assistant, I am Matt's assistant. But yeah, I I don't think this is gonna persist very long. If I'm if I'm running a customer service department,

### [Matt Graham · Passage 40](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-40)

Yeah. I I've tried with my voice agents to call these banks and credit card companies and other service businesses and they just hang up on you.

### [Kevin Williams · Passage 41](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-41)

Yeah, have you tried it? They hang up?

### [Matt Graham · Passage 42](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-42)

Yeah, yeah, yeah.

### [Kevin Williams · Passage 43](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-43)

Okay.

### [Matt Graham · Passage 44](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-44)

Not with Muse, sorry, with the my other ones, but yeah.

### [Kevin Williams · Passage 45](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-45)

So yeah, yeah, I mean it it doesn't surprise me. So open claw could do that too, be able to make calls. But you know, that you have to think that there are two sides to this. In any case, you can do a lot with this little critter. in the space of a few hours, you know, I I I did a lot of the things that it was suggesting. I had it search 50 states worth of databases for unclaimed property, and you know, I turn out to have the John Smith name of our generation. So Sadly, I did not find any like large piles of of dollar bills under any state filings, but it went through 50 states given my address and my history and and tried to figure that out for me. And it it was this really like friendly experience that I mean it was it was fun. I I was frustrated with it because I'm you know I'm a power user, right? I'm like It's not it can't connect to Slack, it can't connect to HubSpot, it can't connect to ClickUp, it can't connect to this, it can't connect to that. Well it's a biz it's not a business tool, it's a consumer tool. What it can connect to is like Instacart. And you know, i yeah.

### [Matt Graham · Passage 46](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-46)

See, I don't know, Kevin. I guess I'm gonna play a little devil's advocate. You know how I like to switch what hats I'm wearing here. I guess I just don't think I I think they're having their moment. There's a lot of excitement, and there's ops obviously gonna be massive value creation for in the consumer market for agents, but I think the there's so much more enormous value and capability within the business environment. So, you know, like I I haven't downloaded Muse because of this exact reason because I'm much, much more interested in the business utility of these agents. And also like I can use my current agents to help me with my Instacart because again, it's like computer use. I have my stop and shop or Instacart or whatever, DoorDash. It's all hooked up. Like just log into my browser and go do what you gotta do if you need it. so I just I think like consumers are curious. Everyone's excited about agents. I just don't feel like the utility is going to be and the value creation will be that significant after this. Two week or three weeks news cycle. I think the most of the agenda conversation will shift back to, you know, business use cases. And that's where we'll focus. And it's still, you know, we still have a ways to go. Like we've said, we've moved from dumb intern to like really good intern to, you know, we're not a full employee yet. we're working up that ladder still. And I don't think like until it's, you know, really good employee level, that's when the it'll take fire in the consumer market. Cause the consumer market's just so much less technical and less forgiving and these extremely easily usability and approachability features, you know. so I I think the hardware side is interesting for fate for meta. I think that's where they have some some interesting stuff with the charm and the the meta glasses, you know, maybe not right this instance, but like what they're trying to build with with connecting AI to hardware in your real life. I think that has a lot of potential. But for me, Muse right now is a little bit on the hype train. We're going to stay focused on Chat GPT work, claude work, cloud co work, and Grochbot I think is where it's interesting.

### [Kevin Williams · Passage 47](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-47)

I I I I think that's totally fair. and I I couldn't after spending a couple hours with it, I it it actually just popped up on my screen, by the way. It's sort of funny. I'm not quite sure what it's

### [Matt Graham · Passage 48](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-48)

Amused it.

### [Kevin Williams · Passage 49](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-49)

doing. it's listening to us and it's listening to us being critical. So that could be a problem for my Amex account later today. Yeah. I d by the way, I I didn't

### [Matt Graham · Passage 50](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-50)

Kevin, stay with me. Kevin, did you get the beeper?

### [Kevin Williams · Passage 51](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-51)

give it access to Amex. I just thought to myself, would I hand Mark Zuckerberg my transaction history on Amex? No. Abso freaking lutely not. And they're they're Yeah.

### [Matt Graham · Passage 52](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-52)

Well that that brings up a whole nother thing about like the privacy stuff. You know, people are worried that there's a lot of conversation about what they want to hook in.

### [Kevin Williams · Passage 53](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-53)

So it so at least according to views, it says, no, in the terms of service, we're not gonna suck in your data, we're just gonna use it. And I'm thinking, yeah, I still don't trust Facebook. Like I just, you know, they're going to monetize

### [Matt Graham · Passage 54](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-54)

Yeah.

### [Kevin Williams · Passage 55](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-55)

this in some way. And monetizing, you know, 25 years of my Amex history is not something I just want to like casually give away. So your mileage may vary. But my issue is is really the habituation of these tools. Like, okay, it did some kind of cool things, and it actually did a couple of cool things in a better way. than some of my existing systems. you know, parents. We we always, you know, your kids are little, but mine are in high school, and you just get deluged by like different course communications through Canvas and trying to figure out this and what has to be done. And, you know, we're not like super tiger parents, but like, you know, trying to be aware of what's going on. It made mincemeat out of Canvas and was able to just log in. And I think that's what it's doing now. It's taking a peek at Canvas and just like seeing what's going on in my kids' courses. And if anything is like, you know, overdue or whatever, it might send me a note. And okay, that that's pretty useful. But I feel like, and this is an issue with agents in general. I don't have that many regimented tasks in my life, such that building either a big clawed code. Codex type project or even having a grok bot. Like, I don't do the same thing enough that those repeatable actions are really worth investing in. But, but, but a lot of people do. And you know, my background is in direct-to-consumer e-commerce and like imagine all of the stuff that happens behind the scenes in an e-commerce company. You've got Customer service orders, you've got shipping stuff, you have or you have like procurement and freight and tariffs, and there's just all of this stuff that's happening behind the scenes. And I do find it interesting that that that people are bifurcating between I'm going to build this really cool like like system based on superbase that's gonna pull in all of this data. I'm gonna have like Not a data lake, but like a data pond where I can like tap into it and I can do all of this stuff, and that's going to be really valuable to me. And then on the other side of the fence is this stuff is changing constantly. And if I can just have Grokbot like log into my warehouse software or my ship's shipping software and basically figure it out on the fly, like imagine your shipping software does a does an update. And between yesterday and today, it's different. If you had like built that system to do something very specific with an MCP connection to the warehouse software, it might very well break. And now you're going to have to maintain that system. Whereas Grockbot is just going to roll with it. And it it, if it encounters an issue, it's going to come back to you and be like, hey, I encountered an issue with the warehouse software. I can't really figure out what the deal is. Can you help me out? That's pretty remarkable. And even as I talk through it, I think I'm starting to buy into there there being a lot of value there, even though almost all of my efforts have been on the like building out the systems side.

### [Matt Graham · Passage 56](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-56)

Yeah, I think so. What I feel is like, let's say in February when you were setting up your open claw, like it was not easy to set up. It took a and again, you have to like s pay constant attention to it.

### [Kevin Williams · Passage 57](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-57)

Mm-hmm.

### [Matt Graham · Passage 58](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-58)

So these tasks, the amount of effort to set this stuff up now has dropped drastically. We've talked extensively about that. And so now suddenly it's worth it. Like if I have a bunch of small twenty minute tasks that I normally do in my business, like the and there's a wide range of them. Now it kind of makes sense to spend a few hours setting up everything for myself

### [Kevin Williams · Passage 59](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-59)

Mm.

### [Matt Graham · Passage 60](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-60)

versus before it would take 40 hours to set up and it would still get stuck all the time. and you would need constant maintenance. So I think people can ask themselves where they are on that curve. Like, are you ready to put five or 10 hours in? Okay. Then you're, you know, and you know, you'll have to maintain this stuff a little bit. There'll be little glitches. You'll ask you, hey, I'm stuck. Can you help me? And you gotta be okay with that. But if you're waiting, if you're hoping like you can just swoop in on, you know, one Saturday for an hour in between lunch and, you know, the football game, get it all set up and then be off to the races, it's still not quite there yet.

### [Kevin Williams · Passage 61](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-61)

It's not quite there yet. And this we hear this from clients too. They like, you know, it's not that it's it's not that I have the luxury to do this, but it's my job to do this, to to

### [Matt Graham · Passage 62](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-62)

Yeah.

### [Kevin Williams · Passage 63](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-63)

like do this tinkering and figure it out on their behalf. But even having to take, you know, a half an hour or an hour and like yesterday it was a client who was they were trying to figure out some nuances in cloud design. They were trying to do something pretty cool in cloud design and you know Doing that tinkering on top of everything else, I hear this quote all the time is like, I'm too busy to invest the time to save myself time or some variation on that, which does make me die inside a little bit. Like just you gotta, you gotta like muscle through

### [Matt Graham · Passage 64](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-64)

Mm-hmm.

### [Kevin Williams · Passage 65](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-65)

some of this so that you can automate it. But I also want to give you like a case in point. Like this is literally happening on my screen right now. And I think it's interesting because it also ties into sort of cost models and et cetera. It's it's a gentic, but it's not it's not grokbot specifically. So I sadly ran out of codex tokens. And I also doubly sadly, as p listeners of the show know, I was sort of slow to the codex chain. I was tied into my buddy Claude. Then I started cheating, but I never had I feel like I'm I'm like confessing something like a dire sin, but I never had a $200 like pro plan on. On Kodaks. They have to be a personal account. So this is a takeaway for the audience. So I've I have my organization runs on both Claude Teams and GPT teams. Again, remember what we do, not everybody needs to do that, but we do. But the business accounts for GPT, you can upgrade a single seat for $100 to a premium seat, but it only gives you a 5x multiple on your tokens. The only way that you can get the 20x multiple on tokens was to pay $200 a month on a personal pro plan. And what a lot of people do, and you know, not saying to do this, but do this, is to have a bunch of a bunch of personal accounts. And if you're a power user, you'd have three or four personal accounts at $200, and then you switch between them. This is sort of a known thing. But GPT put the kibosh on that. And they closed the ability to open new two hundred dollar accounts. So when I l ran out of tokens, I could not do anything. And it I had three days where I didn't have tokens. And it was this this sounds so silly for people who aren't enmeshed in this, but I was just dying inside because this stuff is crazy expensive to use on a per token basis. So I just basically couldn't use it. So I was

### [Matt Graham · Passage 66](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-66)

Did you redo all your resets? Did you use all your resets? You can like Kevin.

### [Kevin Williams · Passage 67](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-67)

I used all my resets. I used all my resets. and well, and then I got clo okay. So so both GPT and now Claude issue what they call resets. And it's like this promotional thing where you can reset your limit. But man, I'm I'm so cheap. Like my reset popped in in the middle of the day yesterday. And I knew that I wasn't going to be able to take full advantage of it before the real reset happened at 6 p.m. So like I actually have one reset stored up. That's, you know, my plan for the afternoon is to blow through that reset. So I spent a ton of time with Claude Code again. And I was doing some really big work and a lot of things. And what I realized is Claude was going to make me work for it in a way that Codex just wouldn't. And what I mean by that is, well, Claude could build out the so we built a new a new product offering is what I was working on. So we're doing 2027 budget advisory. So if you're if you're trying to figure out how the heck you budget for this stuff, we realize that there's a giant gap in understanding. And we're helping organizations like structure their 2027 AI budget in a bunch of ways. So if you're interested in that,

### [Matt Graham · Passage 68](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-68)

Love that.

### [Kevin Williams · Passage 69](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-69)

by all means give me a call. But in any case, when you launch a new product in any organization, think about all the stuff that goes into it. You've got landing page, you've got funnels, you've got ads, you've got all of these things. And Claude Code did a great job on building out the landing page and building out like a marketing language. It helped draft a blog, et cetera. It was able even able to publish it. But through a few hours of working with it, I accumulated a bunch of things that it just wasn't going to be able to do. most of those were like computer control access type things, like how you know, let's configure the campaigns in Google AdWords. And I know full well that Codecs is going to make mincemeat out of it. In fact, that's what Codecs is doing for me right now is Codex is taking Claude's work that it accomplished yesterday of all the things that need to be done, and it's actioning it by logging in with my permission and restrictions to it looks like we've got Google Tag Manager, Google Analytics, Google Search Console, Google AdWords. So it's logged into all of those, plus it's logged into Slack and it's logged into HubSpot. And it is literally doing all of the manual configurations that I would have had to do previously by hand. And behaviorally, I'm I'm like, I just say no. I'm like, nope, nope, sorry, Claudius. you want me to go do something in Google's Search Console? Nope, not gonna do that. So I just wait for Codex and queue off like a giant task in Codex and off it goes. And it is remarkable, like absol freaking lutely remarkable, how good it is.

### [Matt Graham · Passage 70](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-70)

That's what I've saying, Kevin. That's what I've been saving. I'm really impressed with Astro with Codex. Like it's computer use for me is a huge unlock, not to get stuck constantly, not constantly asking me to go click a button at my laptop. I'm really, really impressed. And that's why I encourage people who are just haven't gotten into the agentic game yet for business. This is the first model I think is worth looking into if you're a little further on that curve. Yeah.

### [Kevin Williams · Passage 71](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-71)

But but but so I mean the the you were we're not supposed to talk about models and things, but we do talk about costs. And so open AI just dropped Luna 6. So you know the the obviously the anthropic model naming is all based on literature. You've got haiku, sonnets, opus, and fable. And then on the open AI side, you have Luna, which is generally the lower one, soul, and then Astra above that. Well, Luna six just dropped with computer controls, and it is forty times less expensive on a token basis than than Fable is on the other side. So what's actually driving my computer right now is Luna, not Astra. So I'm I'm this is a little bit of an experiment to see how well that goes. But at least the numbers and the early receipt of it are that, you know. Essentially, okay, cutting to the chase. we have been wringing our hands for months. A lot of people have been wringing their hands about cost, and cost is rising. Now we're like blithely like, Yeah, I have to maintain five two hundred dollar a month accounts and yada yada yada. Okay, that's all of a thousand dollars a month in order to have they're have

### [Matt Graham · Passage 72](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-72)

they're heavily subsidized, remember, they're all training on your data, right? Like they they all are taking your well, I mean, if you're on enterprise you can like opt out of it, but like

### [Kevin Williams · Passage 73](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-73)

No, no, no, no, no, no, no, no. Don't say that. It they're not training on your data, like unless unless you are on a free account or unless you have deliberately opted in. they that's that's a big deal. in any case, so with this level of intelligence, if this actually plays out the way that it's supposed to, if Luna works nearly as well as Astra at Like, you know, it's like eight times cheaper than Astra. like we're kind of good. That's gonna this this this is what answers the subsidy question is it the models keep getting more efficient. And the latest and greatest models that have been launched are as good at 40x decrease in cost. That means that even as the subsidy goes away. we should be able to get comparable performance for near comparable cost. That'd be my

### [Matt Graham · Passage 74](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-74)

Definitely they have to. I mean, so this is like kind of a another argument of does it really matter if the models get significantly better if they're already good enough? You know, they're already good enough at so many tasks. Like they're

### [Kevin Williams · Passage 75](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-75)

Absolutely.

### [Matt Graham · Passage 76](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-76)

still not good enough to do, you know, like we keep talking the agentic AI, the computer use, being able to do these long horizon tasks like an employee. They're still at good intern level. so there's definitely still room for improvement. But for other types of tasks, writing content, whatever it may be, it feels like they're plenty good enough. Doing research, writing reports, generating some some types of images. yeah. I I just I feel like we're already ninety five, ninety-seven percent of the way there on certain types of use cases.

### [Kevin Williams · Passage 77](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-77)

I I think this has been observed a lot recently. Ethan Mollock, who's always one of my favorite AI influencers, just posted about this a couple of days ago. That okay, we talk about you know, recursive self-improvement and AI killing us all. And y it took us all of a week to move on beyond that, by the way. Now we're back to woohoo, this stuff's awesome. yeah. but i the observation that r from a regulatory perspective, something is eventually going to happen. But We have all we need to to completely transform pretty much all business, probably all humanity, given the technology where it sits today. And I don't think it's gonna stop quite yet, but like we're pretty set in that respect right now. So I'm not overly worried about it. I would push back about the smart intern comment. Because what I just described as far as setting up AdWords, getting into cons search console. I mean, it even went into cloud console. I think about skilled marketing people that I've had on my teams, that is a week's worth of work, easily a week's worth of work. And it did it in four hours as I was having dinner. Like that's not intern work. That's actually pretty skilled technical marketer work, which is a which is a which is a level of capability up.

### [Matt Graham · Passage 78](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-78)

Yeah, okay. I I like to be facetious. I think still there's a wide slot swath of tasks and workflows and jobs that employees do that it's still good intern level. But yes, there's definitely some use cases where, you know, like video editing is really on the rise. I think it's really doing well in that area too. You know, all these different types of media and marketing jobs, I think it's getting pretty good at, definitely. I just saw a hilarious video. this influencer who the CEO jumped on a call and you know, there's like a four person marketing team and he's like, Hey guys, you know, this AI stuff, I really think we need to slim down the team. And he just inter he just like interviews the person right there on this like four person team. And then he asks like Fable on the side, like, Hey Fable, do you think you could do that? Okay, okay, just

### [Kevin Williams · Passage 79](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-79)

Mm.

### [Matt Graham · Passage 80](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-80)

go write the termination letter. And then he removes the person from the call and then boop, all right, we're good, we're good. We like and he it's is

### [Kevin Williams · Passage 81](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-81)

It's funny until it's not.

### [Matt Graham · Passage 82](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-82)

Yeah, it's like it's meant to be, you know, it its message is scary, but it does it in a funny way. That's why I'm laughing. But anyways, my point is I think it's making a lot of head headway in the marketing area. For sure. Yeah. So

### [Kevin Williams · Passage 83](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-83)

Yeah, yeah. All over the place for sure. I mean yeah.

### [Matt Graham · Passage 84](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-84)

so Kevin, another interesting topic here is how these companies are buying data. So we've talked a for

### [Kevin Williams · Passage 85](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-85)

Yeah.

### [Matt Graham · Passage 86](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-86)

a long time, like, you know, the barrier for agentic AI has And computer use has been having the right kind of data to train these models, right? And we're seeing these models get better and better. And, you know, in some areas they have more room to run and to improve. And I still feel like a fundamental barrier for them is actually having the right data to train on, right? Like, you know, we have a lot of code in the world. That's why we could crack coding. We had a lot of text in the world. That's why we were able to crack generative text, computer use, agentic AI. Taking a little longer. my gosh, it's been, you know, like nine months and we're still not there yet, Kevin. But you know, the data, the fundamental data that it trains on has, I think, been somewhat prohibitive for it to improve and even get better and better. So, you know, I've seen a lot of companies getting approached about selling their corpus, corpus, their entire data set, which think of it as like a Slack export, Google, if you use Google or Microsoft exporting your entire workspace. And allowing them to repackage that and sell it to the frontier models. So I've personally had tons of calls with these companies just out of curiosity to see what it's like. It's a fascinating space. I know everyone probably remembers the headline of Google and a company called Micro One getting into a bidding war to buy Spirit Airlines data. But there's a lot of like that Micro One company and a lot of other vendors out there who are trying to buy, you know, businesses that just have over 25 people in America. Working for them, they might pay you hundreds of thousands of dollars for your data.

### [Kevin Williams · Passage 87](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-87)

So so let's I I think the Spirit Airways one is really assailable as far as understanding it. This was just a few months ago. And so RIP Spirit Airlines went south, right? So Google ends up paying $10 million for not planes, not not customers. Well, actually, I guess technically they bought customers. But for the operating data under the hood, and it's boring stuff. It is the operating documents, it is the Slack messages, it is the emails, it is you know, hiring logs, et cetera, et cetera, et cetera. Why, why, why, why would they do this? Okay. Just from a raw data cut, it's interesting. It's it's, you know, terabytes of information. Espirit was small, but it wasn't that small. So the amount of data that's in an organization that has a few thousand employees and you know, yada, yada, yada, you just just using that for training is interesting. But what it sounds like they're doing is they're actually building a sandbox digital airline. So now you have airline in a box that has real data, and now you can start experimenting on it and doing testing and trying to figure out. How AI could replicate the behavior of airlinus generalis at the end of the day. And at the end of the day, they are going to end up with a really good model for airline or transportation operations that they can be overlaid into the operations of a real airline, right? And it's not like, hey, American or Delta or whoever. I I'm I'm sure they're all working on this, but let's figure this out. Let's see what data sets we have. No, no, no, no, no, no, no. Google can come to the table with a data set that's already attuned to what the business does. So it's like they're starting on third base as far as being able to automate that. And you know, other versions of this are like Mercor, which we have talked about on the show before, which is paying hundreds of dollars an hour for senior professionals to basically do human reinforcement training. on being a controller or being a law firm associate or whatever it is, all of those decisions are being baked into the next generations of models, which is going to shorten the timeline of automation in all of those industries.

### [Matt Graham · Passage 88](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-88)

Yeah, it's super interesting. Scale.ai is another big one. It got bop. I've

### [Kevin Williams · Passage 89](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-89)

Yeah.

### [Matt Graham · Passage 90](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-90)

met a long time ago. But same idea. Training having humans give the thumbs up, thumbs down

### [Kevin Williams · Passage 91](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-91)

Mm.

### [Matt Graham · Passage 92](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-92)

is the simple version of it as they train the model and do that reinforcement learning with the human in the loop. But you know, like a new model just came out. You know, here I am on Friday thinking like, okay, we're good, Kevin. We got through the week. There's no apocalypse, you know, the machines didn't kill us, no new news. I can rest easy. I look at my phone on Friday night and there's this new thing, Geeve. That just came out. Okay, great. No and it's like flipping reinforcement learning on its head. Have you dived into that at all?

### [Kevin Williams · Passage 93](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-93)

Is so J E V, I'm not is it is it Jeev? Is it Jev? I'm not I'm not actually sure. I I don't know. it's a problem. This is the problem with with

### [Matt Graham · Passage 94](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-94)

I think it's Jeff. We know it's so new we don't even know.

### [Kevin Williams · Passage 95](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-95)

I'm I'm either reading my information or I'm having

### [Matt Graham · Passage 96](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-96)

Yeah.

### [Kevin Williams · Passage 97](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-97)

AI give it to me, and I'm I'm not really sure what the pronunciation is. So I apologize to the Jeev Jev people. It's not yeah, anyway. okay, why is this important? So this came from one of the original founders of OpenAI. And you know, large language models hoover up all of this information, spirit airlines, the whole internet, et cetera, and they make connections between all of this data. But it is a ponderous data data set. It is absolutely massive. And that is where a lot of the costs are born. Is you know, one of the first unlocks we had was actually last year or two year and a half ago with Deep Seek, which which Really came up with some technology that's called experts architecture that allows different areas of the neural net to activate. So if you want to talk about accounting, when you and I talk about like context and prompting, that's this is sort of why you need good prompting still. Like telling your LLM that it's supposed to act as an accountant activates the part of its neural net that is based around accounting and allows it to execute more efficiently and get you better answers. Okay, fine. But it's still super ponderous and super expensive for it to go through its giant hive mind. So what Jev, Jeev does is it's a different, it's not a, it's not a large language model. It's a basically it's a decision-oriented model that is deciding in almost like a flowchart or decision tree type way, very, very rapidly whether it's going to be path A or path B. And it's like, is consumer happy? Yes or no, based on sentiment variables. And if happy, now we're activating a certain area of our decision tree. If unhappy, we're activating a different area of our decision tree. And what that does is it routes the LLM decision-making authority or ability to the right area of the neural net much, much, much faster and much more efficiently. So, you know, they're claiming like, you know, 250% reduction in cost. Now, now, there's a lot of caveats here, like you have to have the right use case and the data has to be formatted in the right way. But I think this is really, really interesting because it's just another building block of the technology that allows these things to happen faster, betterer, cheaper at the end of the day, that will be layered into the technology over time. Can JEV work alone? I mean, there are some use cases we're sure it can. Can JEV work in conjunction with a large language model? yeah, absolutely it can. And although it it it needs some more time to ripen, already you can see the potential. And then it ladders up from there. Like, okay, we've done JEV, we've done decision models, we've done large language models. Then it's the world models that we talked about a couple of weeks ago, as far as physics and things like that. And all of these technologies are start compounding on top of each other, which is why at the end of the day, long-winded way of saying, I am not as worried about the cost apocalypse as a lot of people are.

### [Matt Graham · Passage 98](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-98)

I'm definitely not that worried. I think the if we keep having breakthroughs at this rate, I I I think it could get confusing maybe within ChatGPT or Claude of like which model do I use and really understand what the true cost is. But I think all these will get really s proficient at routing and switching. and I think that plays back to the data center too. I know we're not supposed to get into all this stuff, but You know, data centers

### [Kevin Williams · Passage 99](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-99)

Yeah, we do.

### [Matt Graham · Passage 100](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-100)

are really expensive and we're building out a lot of capacity. But if these models can get really good as well, you know,

### [Kevin Williams · Passage 101](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-101)

Mm-hmm.

### [Matt Graham · Passage 102](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-102)

that capacity can go a lot, lot further than than what what what people are saying will be the, you know, the re rate of return. Everyone's worried the bubble will pop because of the rate return and all that stuff. so yeah, I think it's really encouraging to see a fundamental architectural change like this happen in our country, especially, and also attack not just, you know, a intelligence at the higher end, but just start at attach attacking speed and cost. And I mean, these are huge, exponentially better numbers. Really interesting.

### [Kevin Williams · Passage 103](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-103)

Yeah. Yeah, it's i it is interesting. I mean I I get a little frustrated with the investment environment and the bubble and like all that's going on there because I actually do believe that the demand conversation and the supply conversation are not as interlinked as they need to be. That the that the supply people who are the data center people have one set of models and whether those models actually align with demand reality as far as willingness to pay, ability to absorb, like it's really unclear to me. I mean I'm actually been invited on a on a on a on a d a supply side podcast next week to talk this through with some leadership on that side because you know they're open eyes about the fact that they don't really understand it. And guys like us have a pretty good tap for how this works. And like I speaking for my own clients, Like you really couldn't quadruple the costs involved and have them lean into a lot of the projects that they're pursuing now. it wouldn't make the same level of economic sense, balanced with risk, balanced with understanding, even though a lot of this stuff's just amazing. Like they're just not going to pay four times as much. And if that's the case, then you have a you have a problem on the supply side as far as the demand willing to pay. Even though demand keeps going up, which is good, but does it go up at the same steep slope that the those people are anticipating?

### [Matt Graham · Passage 104](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-104)

Well, that's the bil trillion dollar question. Nobody knows. Trillion dollar question.

### [Kevin Williams · Passage 105](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-105)

Trillion that's the trillion dollar question. The other one, so as we c sort of close out here, that's the trillion dollar question is the you know, the IPOs of both open AI and anthropic. And you know, there's a some animated conversations going on about the leapfrogging that open AI is doing right now with things like Astra and whether or not Anthropic's ever going to be able to catch up. And just the sort of this just

### [Matt Graham · Passage 106](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-106)

I don't know, Kevin. What do you think?

### [Kevin Williams · Passage 107](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-107)

just so so so I am I'm I I'm still an anthropic fan. I believe in their technology, I believe in their teams, but there is a little bit of a cold reality that OpenAI has much more process that has that they have availability of. Anthropics suffered mightily from the inference demand they had when they started to basically win the wars at the beginning of 2026, which meant that they spent fewer tokens training new models and more tokens on inference. And the question, and it is a question, I don't have an answer for it, but this is this is factual is if OpenAI has been investing much more technology capacity in their new model development, will anthropic be able to catch up when they haven't been doing the training spend that OpenAI has? And if we get to Astra 7 before we get to something comparable out of anthropic, that's going to be a real problem for.

### [Matt Graham · Passage 108](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-108)

Yeah, it's interesting. This like I I d do you believe in this theory that whoever hits like, you know Yeah,

### [Kevin Williams · Passage 109](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-109)

Recursive self improvement.

### [Matt Graham · Passage 110](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-110)

like then they're gonna have a takeoff

### [Kevin Williams · Passage 111](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-111)

Yeah.

### [Matt Graham · Passage 112](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-112)

speed and no one will catch them. You know, like I I understand it like logically and I don't really have any logical counter argument, but it just still feels not intuitive to me that that will happen.

### [Kevin Williams · Passage 113](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-113)

I mean, I'm I'm feeling it a little bit already. Like if I had my druthers and I had the ability to have my $200 plans, I would probably be running three two hundred dollar codex plans right now. And I wouldn't be messing with anthropic more than as like a double check. And they're and so Claude Design is great. That's still like the best design tool that I know of that's out there. there's certain things that I prefer Anthropic for from a user perspective, but If I could run all day, every day in some version of Astra or something that has Astra capabilities, even if if that's Luna, 100% I would do that. because it's just so much better. So the

### [Matt Graham · Passage 114](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-114)

Yeah.

### [Kevin Williams · Passage 115](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-115)

question is, does Claude, whatever the next is, catch up to a degree that I can balance that out? Or does user preference start shifting in the open AI way? So It all shifted from OpenAI to Claude. People are fickle and they can shift back and forth. There's really no barrier to switching. so I I don't really know to to to to put it succinctly.

### [Matt Graham · Passage 116](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-116)

Well, that's a great way to end it. We have left with more questions than answers.

### [Kevin Williams · Passage 117](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-117)

As always. All right. Well, if anybody's e interested to try out Muse, they sent me like an affiliate code thing. so they will give they'll give you a billion credits on it. Who knows what a billion Muse credits actually means, but it sounds like a lot. So I'll put a link in the show notes on that. full disclosure, I think they'll give me a billion tokens that I have no idea what I'm gonna use for. So My little muse bot can go out and torture my children by looking into their grades on an hourly basis instead of on a on a daily basis. and yeah, until next week, sir.

### [Matt Graham · Passage 118](https://ascendlabs.ai/podcast/ep-43-is-the-cost-apocalypse-already-over/#transcript-118)

Yeah, till next week. Thanks, Kevin.

