# EP #22 When Your Entire AI Team Calls in Sick at 6 AM

“EP #22 When Your Entire AI Team Calls in Sick at 6 AM.” AI at Work by AscendAI, 2026-04-16. https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/

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/).

### [Elijah Szasz · Passage 1](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-1)

What's up, Kevin? Looks like you're on the road because this time it's your turn to be strapped into a $12 pair of Apple AirPods that are wired like me last week. No fancy setup just in case your audio doesn't sound super pro today.

### [Kevin Williams · Passage 2](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-2)

Hey man, how's it going? Yes, indeed. visiting your Yes, although it should be all right. Well, I'm visiting your old town. I'm in LA. I'm in Santa Monica.

### [Elijah Szasz · Passage 3](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-3)

Yes, you're right. You're in Lala. ⁓ you're in Santa Monica. Like we're in Santa Monica. you?

### [Kevin Williams · Passage 4](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-4)

nowhere I'd like to advertise. ⁓

### [Elijah Szasz · Passage 5](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-5)

Okay, you sure about that? Send a drone there. will right after this send you a long list of places you have to go eat and then tell me how much you love them and I miss them so much being in a little bit of a food desert here in Utah. But I'll give you some good recommendations. Kevin, happy tax day. The thing that makes tax day a little less enjoyable. is having every entropic product go down that morning as I'm trying to finish my filing. And I've got like these massive projects with tons and tons of attachments doing this assessment and then doing a last sanity check before I sign and send to the IRS and then give them all of my money. And as I tried to do that, I got a smattering of all sorts of little error pop-ups that ranged from Spanner not found to stripe not connected. And I'm like, I'm not using Stripe. What are you talking about? Just, just, all these errors. And ⁓ course I go to ⁓ status page and it's just lit up red across the board. and ⁓ I through my little usual routine of trying to work around it. Well, maybe the desktop app works. Maybe the mobile app works. And then just gave up, but it's back up and running. It's. crazy. sent you a text message and I said, I had this moment where felt like I walked into the office ⁓ morning and just the entire team was out sick, like everybody, the whole office was out sick. So I had all these agents were supposed to be running all these things I was supposed to do. And I already started thinking ahead about how much of my day was completely nerfed because this one model took a crap. It's wild, right?

### [Kevin Williams · Passage 6](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-6)

So in my case, I woke up at like 4 a.m. because I'm on Pacific time and I was so excited to use that time to like, I was gonna build this and I was gonna build that. And I had all of these plans and I got just deep enough into it. It broke at about 6.15 this morning that I was totally stuck in exactly the same way. So when you sent me your text, what did you say? Something about realizing that you were dependent on one dealer and.

### [Elijah Szasz · Passage 7](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-7)

get shit done. ⁓ I said I it's like I woke up realizing that I've got this awful heroin addiction and never made a plan of what to do if my dealer disappeared.

### [Kevin Williams · Passage 8](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-8)

And yes, so what did I do? I trotted around the corner and I started whispering to my friend Codex over at GPT land. I cheated on Claudius and I started actually leaning into trying to figure out how to create some duplicate versions of my workflows. A, because I've kind of been meaning to anyway, just to get a better feel for where Codex sits. we've been talking a ton about Claude Code on this show over the last weeks.

### [Elijah Szasz · Passage 9](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-9)

Yeah. Yeah.

### [Kevin Williams · Passage 10](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-10)

because it is sort of getting the edge. But these instabilities are causing people to experiment more with Codex. And if you're in the GPT world, it's not like OpenAI isn't paying attention to what Claude is doing. So there's a lot of functionality that is popping up on the GPT side to replicate Claude co-work. Codex is faster, it is cheaper to run, and often it will catch bugs that... that cloud code won't. I'm not necessarily jumping ship or anything like that, ⁓ understanding that there are other options out there is pretty important for resiliency in your workflows and resiliency in your business. The tool thing, however, that is a problem. I think we have 22 different that we've built over the last few months that do different things in the business ⁓ that from workflows to actual platforms like the platform that we're going to use to process this podcast, for example. It's all built on cloud. it really isn't ⁓ yet make a platform that does automatic model shifting because the underlying logic changes and the prompt changes and it's pretty challenging. ⁓ So have a single source of supply ⁓ and where it's... it feel like everybody's out sick today because all of the tools broke at once.

### [Elijah Szasz · Passage 11](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-11)

And not to throw this word around loosely, but in agentic workflows, when you're talking about some of these tools that you're calling, and you have a whole series of steps, those steps might be very specific to how that model works and handles those steps. And you might've gone through weeks or months of iterations on that, like I have to get it to where it is. And you can't just sub out the brain that's running all that seamlessly. Like you can't just say, Oh, hmm, clods down. I'm just going to. hop Gemini into this and expect it to work the same. And I really think it's worth mentioning and maybe unpacking the difference because I want to get into not only that redundancy, but the costs associated with powering these things. The difference between this web UI, like this browser that most people go to, most people aren't even paying for this experience, right? They just go to Anthropics website and they type in a prompt and they get an answer they have this experience or maybe they're paying 20 bucks a month so they can have better models and more usage, or maybe they're like us paying either 100 or $200 a month to get even more out of that and more usage and longer session durations, right? And that's one experience, but the experience that you're talking about with the tools that you built are actually using the API key in your account. And that has absolutely nothing to do with that $2,000 or $200 a month that you're paying. It's completely different. And I want people to understand that because this is really, really important for how it's being used at work and important because of some changes that Anthropic made 10 days ago to how this all works. It's very meaningful, ⁓ financially least. So yeah, do you want to take a first pass at that, Kevin, just kind of differentiating how those two worlds work independently and sometimes coexist for people?

### [Kevin Williams · Passage 12](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-12)

I think it's a good point to be made that as you said, most people are just signing up at $20. using a max plan on Claude, it's $100. GPT now offers a similar plan. And you're using either the browser functionality or if you're using cowork or if you're using Claude code or codex on the GPT side through the apps, ⁓ you have a number of tokens. ⁓ everybody knows what tokens are at this point, but just it's the base unit of currency of AI. We hear a lot about this. It's about 75 % of a word, roll with it. tokens are the variable cost of this stuff. And if you're paying for the $100 max plan, that gives you an allocation of tokens. by hours and hour blocks and you will exhaust more tokens by using a smarter model. Like the Opus model is smarter than the Sonithion model. It thinks longer, it uses more tokens. If you have really long context, like you're using a really long chat, recognize that by the time you get to the 300th... line in that chat, it's actually going back and compounding all of that information that's behind it and will rapidly use up your tokens. So particularly on the cheap club plan, if you use Opus, you can run out of tokens in like six or seven prompts or even three or four prompts if you're doing something complicated. And it'll flash up this warning that's like this, your quota will refresh in four hours and your like your hands are hovering over the keyboard. And it does give you this handy dandy option to pay as you go, which is not a bad option, but it can be really, really expensive to use. So sort of there's a side side note pro tip there is put $50 or a hundred dollars into Claude so that you don't so totally grind to a halt. So you can actually finish whatever that that process is, but don't rely on it because it's going to be really expensive. Okay. So that's, that's the paid universe. That's just the plan universe of how these things work.

### [Elijah Szasz · Passage 13](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-13)

Yeah.

### [Kevin Williams · Passage 14](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-14)

And then on the API side, this is where the real dollars are being spent. Where if you are developing an app, okay, so the app that we built to do ⁓ podcast processing ⁓ ⁓ has nine or 10 different API calls to Claude to do certain things in it. So we built ⁓ the ability to ask. all of the transcripts of the podcast questions. So in case we're on the pod and we're not sure if we said something before, we can really quickly ask a question of basically our own private chat bot. Well, that bounces off of the information that's in the database and it goes to through the API. And then we're being charged for that based on fractions of a penny per, you know, thousand tokens or million tokens or whatever it is. And that's totally different than it just being your monthly plan. Reasonable?

### [Elijah Szasz · Passage 15](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-15)

Yeah. Yeah, here's what I'll add to that. And this is where the changes really make a difference about what happened 10 days ago with Anthropic. That meter that you're talking about with this API call and those tokens that you burn, it's much more measurable. It's much more of an exact science. You can really start to understand a certain process that your software is doing or a query that somebody is making on your software and then just... how many tokens that is going to use. For example, I had a good friend who did this really cool business analysis app and it was using Opus in the background, which is the big fancy cloud model. And then she realized that every time somebody did this, and this was a lead magnet, by the way, it was costing her $30 in tokens, right? Not very sustainable, but she could measure that and then see that outcome and then plan for it pretty precisely. Now on the consumer side, when you just pay your 20, 100 or 200 a month, it kind of reminds me of how everybody back in the day was promising that Bitcoin was going to be the next thing that we were all going to transact with. I was going to buy my coffee with Bitcoin or a leather jacket with Bitcoin or a motorcycle with Bitcoin, but it changed all the time, right? The actual meter changed all the time. So one day it might be a cup of coffee and the next day it might be a motorcycle, but it's the same amount of Bitcoin. So what's been happening with Anthropic on this consumer side with your regular subscription is that, you know, I'll be working and I'll send a text message to you and say, hey, what's going on? I'm just using Sonnet and all of a sudden I hit my session limit. Yeah, me too. This is so weird. It's like they're throttling my usage. I'm like, I got to keep working. I'm going to buy another $100 worth of credit here. Okay, cool. Yeah, me too. And then all of a sudden, like the next day it's like, Whoa, wait a minute. I only got three prompts out of Opus and now my entire session allocation is gone and I'm paying a hundred a month. What's going on? But then the next day it's different. All of a sudden, next day you get like twice the number, three times the number of prompts. So somebody is back there behind the curtains and they've got these levers that they're pulling and they're constantly adjusting and testing people's price sensitivity and tolerance. And after a exceptionally bad week of running out of all of our token allocation on the paid plans, both you and I noticed that we got these little notifications saying, Hey, claim your free a hundred dollars, your free $200 of credit. Because if you start asking around some of the models to do some research of, I just going crazy or is this a problem right now? It'll go scrub all these Reddit threads and come up with a lot of people making a lot of complaints. Right? That kind of plays into what happened on April 4th. So of the little workarounds to having this broader of tokens on a more of a forgiving meter than the API call was to do things like hook an agent up to your max plan and then just it run wild. And technically you were violating their terms of service by doing so. There were some stories of people's accounts getting suspended, but I still know all lot of people who did it because the alternative was paying hundreds, if not thousands of dollars a month to have these agents run. So on April 4th, Anthropic killed the little workarounds that existed before that thousands of developers had quietly relied on. And that was this ability to run these third party party agent frameworks like OpenClaw, which we've talked times about on the show. I'm not going to go too deep into that one right now. So flat Cloud Pro or Mac subscription. ⁓ had like maybe 135,000 open-claw instances that were running it all the time. you know, the economics of doing that for 200 bucks a month versus like a thousand ⁓ $5,000 worth of agent compute, really pushed Anthropic to patch every pathway with hard errors. ⁓ And you want to run an agent on cloud now or power your app or do whatever you want to do with that, you've got to use the API and you have to pay per token. And Opus 4.6 runs $15 million on input tokens and $75 million on output on the regular. And so a lot of this really a lot of people. offered, Anthropic offered a bunch of credits to ⁓ other monthly plans. They started throwing discounts around because there was a big uproar. But we've always about over and over, Kevin, that we're getting a disproportionate amount of value. from the price of all of these models today, all of the frontier models, that at some point, the is gonna stop, right? Like there'll be a day of reckoning where you have to pay. And I feel like this is one of those dominoes ⁓ that 10 days ago.

### [Kevin Williams · Passage 16](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-16)

Yeah, this is a super active conversation.

### [Elijah Szasz · Passage 17](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-17)

Yeah.

### [Kevin Williams · Passage 18](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-18)

My bet, and maybe I'm wrong, is I thought it was gonna take a little bit longer, cynically they were going to really let us get hooked on their heroin. And start making staffing decisions ⁓ and the way their businesses run, and then you wake up on some Tuesday and they're like, look, this isn't economic for us because we're a nice company. We're not gonna just like, you know, do this blindly. We're going to tell you that we're doing this. because we only have so much capacity and we're draining the lakes in the Midwest for your cat videos and whatever. like we're gonna have to start charging you, but it does feel a little bit premature and just based on my own reaction this morning where I still have plenty of flexibility and choices. It was a little bit painful. But I actually had my workflows, my core personal workflows that link in HubSpot and ClickHop and my email and my calendar and LinkedIn and like all of these different things. It was pretty easy for me to just recreate all of that on the GPT side. ⁓ then you have the gorilla in the room ⁓ is Google who's standing out there who has ⁓ those three, it's the only one that actually has plenty of cash that the other two, okay, they have a lot of investment capital.

### [Elijah Szasz · Passage 19](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-19)

Mm-hmm.

### [Kevin Williams · Passage 20](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-20)

And yes, they are starting to create revenues in the billions, but their expenses are in the hundreds of billions. So Google is still has, it's just a cash cow as far as making all of this money. they all have to sort of play the game of racing towards features ⁓ that you acquire users ⁓ waiting out the market tolerance to pay more for those features. knowing full well that Google can be the spoiler out there and keep their prices low basically and just starve the other two.

### [Elijah Szasz · Passage 21](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-21)

Yeah. Yeah. I mean, I think that's pretty accurate. far as labs shipping stuff, it's just exhausting. ⁓ amount of stuff coming out all the time is exhausting. But again, ⁓ it was a race, Anthropic is winning it. I've been working in tech for 20 years. I've never seen a single company ship so much stuff so often. It just makes your head spin. getting my hands in there and using it for many, workflows ranging from writing to complete automations that are doing the job of two or three different people on a daily basis. I'm still getting the best results and the best utility out of those entropic models. So when they do a price change like this, or they shut off one avenue to be able to use the product, say like, ⁓ now you're going to start paying. It definitely, it definitely hits as a small business. And you know, makes you start to really consider, okay, what are the contingency plans? far do you want to go in one direction before painting yourself into some technology corner and not being able to have alternatives, right? Like it's, AI is a lot of problems right now, but if you're not careful with your planning, with your change management, with how you set up this manifesto in your company, it can get you into hot water really, really quickly because it's a very, very quickly evolving place right now. This is all changing.

### [Kevin Williams · Passage 22](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-22)

So this is also just a point in time though. think the other contention that I make, and I may have said this in a previous episode, 4.6 is amazing. ⁓ if they just stopped everything now and said, you get is Opus 6, people us and people like most small companies, even with some of the foibles and issues, it will absolutely change everything. And I've been saying this for three years, as each model comes along, ⁓ this a really, really good one. And yes, there will be an Opus 5.0.

### [Elijah Szasz · Passage 23](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-23)

Mm-hmm.

### [Kevin Williams · Passage 24](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-24)

point that it's going to be like the mythos thing and like it's going to be absolutely nuts. But if you look at the historical cost of the older models, they get commoditized and they cost less. So right now, $15 per million tokens is the OPUS cost. Yesterday's or today's OPUS is tomorrow's Sonnet. So you'll have Sonnet 5.0, which will only be $2, that is probably the OPUS model of today. And what a lot of us are going to end up doing is we're going to be finding the least common denominator model that meets the business case. I suspect that the platforms will do things like off time ⁓ ⁓ tolling and things like that, ⁓ that, okay, you're using a cheaper model. So the trade off is it's gonna take longer. You don't just get to pivot, pivot in your chair and have your app done. You actually have to go get a coffee and come back to have your app done on the other side or I think about all the automations I've built, most of them trigger at five o'clock in the morning. And that's sort of arbitrary. Like, couldn't they just queue up over the night and take advantage of excess capacity in the models? Sort of like a load controller in a house.

### [Elijah Szasz · Passage 25](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-25)

Well, that's, and that's, that's be fair. That's happening right now with Anthropic too, right? Like that's, that's another thing is that it's kind of like getting an Uber. a 5 PM on a Friday night. Guess what, buddy? Surge ⁓ And same thing is starting to happen right now with peak business hours with using these models. You're actually going to rip through allocation of tokens on these paid plans faster than if you were running something at two or three in the morning. So, you know, now I'm playing this scheduling game, which I actually have an agent doing this for me.

### [Kevin Williams · Passage 26](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-26)

Yeah.

### [Elijah Szasz · Passage 27](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-27)

looking at what jobs can run on what hours so that I'm getting the most out of my plan. Again, as long as I stay inside of that entropic ecosystem and I'm not making API calls, you can really start to stretch out that monthly subscription as well based on when you're using these different tools.

### [Kevin Williams · Passage 28](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-28)

So let's be clear about this. When is the cutoff? So I'm on Pacific time, you're on Mountain time. What is considered prime time now as far as Anthropic? Again, that's that window that you tip into and all of a sudden your tokens get consumed by 4X the rate. So, okay, you wanna stay outside of that if you can. For our listeners, when is that time? Right now, this is probably gonna change, but what is it right now?

### [Elijah Szasz · Passage 29](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-29)

Mm-hmm. Yeah, I think last I looked, it was East Coast time between eight in the morning and 11 in the morning was like considered peak. If you went into the afternoon, it lessened. If you went into the middle of the night or early evening, it actually really good. Middle of the night being the best. again, when I there are ⁓ little dials turned behind the scenes. We still don't know exactly what the numbers on those dials are, how far they're being turned or when they're being turned. So all this stuff is really speculative. And a lot of it comes from just the experience of seeing like, ⁓ wow, just, ⁓ session just went up and I just had like three prompts on Opus what happened. But at o'clock at night, I had nine and I was fine. That's weird. So a lot of it, you just have to get in there and start using this stuff. But did say that all of this kind of time-based metering was stuff that they were rolling out, along with filling of the little holes that allowed people to use the paid plans to ⁓ run agents stuff like that. That's what mean by ⁓ it's ⁓ very This stuff is just changing every single day. So it's gonna have backup plans.

### [Kevin Williams · Passage 30](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-30)

So a little bit of a gut check here that most people who are really upset about this are using the $20 plants and that is so cheap. given that this is this amazing supercomputer that can automate like parts of your work. It's very odd that we've calibrated on this very, very small number. And at $20, you get X. At $100, you get 5X. Okay, that's linear. At $200, you get 20X. So it's giving you a big bump. And now that I'm on the $200 a month plan, I basically don't run out.

### [Elijah Szasz · Passage 31](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-31)

Yeah. Wait, wait, you're back on the $200 a month plan because you went down to the $100 a month plan and now you're that just wasn't enough for you.

### [Kevin Williams · Passage 32](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-32)

Yeah. I keep trying. Well, well, well, okay. I play the game every month. I take, I... I'm this cheap. I go into my clod subscription and every month I downgrade it to a hundred. Okay, cool. And then the next month if I end up in a pinch, I'm like, heck. Okay, now I have to upgrade it and ⁓ upgrade it. And this month I made it all the way to April 1. I had to upgrade it on April 1st. ⁓ So.

### [Elijah Szasz · Passage 33](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-33)

Wow. so I think it, yeah.

### [Kevin Williams · Passage 34](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-34)

Yeah, but $200 a month, $200 a month. Like if you don't know what you're doing and you're just getting in there and you're chatting and you're doing kind of basic stuff, okay, I get it. But as soon as you bridge beyond the chatbot and you start doing an automation, like I would pay $200 a month for most of the automations that I've already built. I hope Dario isn't listening. ⁓ would probably pay $500 a month. And where is that equilibrium in the market ⁓ the value that people are getting? ⁓ I don't think that enough people are getting that value out yet. And it would be too much of a pushback, either based on API pricing, which can be very expensive, or upping these plans. So as I said before, it's just a bit of a waiting game. until that happens and you basically want to be in a place where the value is so clear. It's like shut up and take my money. Cool. 500 bucks. Totally get it. no longer have to do X, Y, and Z in my world ⁓ formally I was having to pay a contractor or a vendor or something to do that. So this is no brainer territory. It's not there yet for most people.

### [Elijah Szasz · Passage 35](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-35)

It's not. And I think when you're so deep in the weeds with this stuff, a lot of it hits different for you, right? the economic impact. But it's always interesting talking to other people and how they're using this at work. And I had a conversation with an agency owner yesterday and really got a sense at how material that economic shift can be with one small rule change. So with what you were talking about, Kevin, when you said, ⁓ I have all these tools. And then when entropic went down today, all those tools stopped working because they're making this API call all of those tools that you're using. You don't really have a choice. Like you built something custom, you used Claude code, and then you got to hook it up to, mean, you could hook it up to whatever model you want, but you hooked it up to either Sonnet or Opus to do what it has to do. and you're just paying per API call. you're paying for the tokens, you can put limits on it in case it gets too crazy, but that's just the world you're in. And so, you know, this open source, ⁓ agentic framework that we've talked about a bunch of times, OpenClaw, what it does is, you know, it's, you can hook it up to whatever model you want, right? You could hook it up to a variety of models. You could do model routing. You could say, I want to use GPT for this. I want to use Gemini for that. I want to use Sonnet for task over here. And it will go and pick and choose and do all that for you. But what a lot of people were doing were paying for that $200 a month max plan with a couple of little tricky workarounds in the terminal. They're able to power these agents with that. Right. So when that stopped, when I, when I talked to this agency owner, he said he had three different open-claw agents that were actively in three different Slack channels, like for clients, for his team. These things are doing the whole work. They're producing content. They're publishing content. They're doing all sorts of stuff. And I was like, wow, that's pretty amazing, man. ⁓ model are you using for these agents? And you said, I'm ⁓ using Opus. mean, these are pretty complex and it requires Opus. Wow. Okay. Okay. And you basically can't hook up your pro plan to What is that looking like for you as far as ⁓ token costs? You want to take a guess like what these three agents are per month? And listen, man, I will preface this thing like this is nothing earth shattering. Like, ⁓ my gosh, I can't believe they're doing this. And you figured it out. It's like, yeah, they're doing some human work. And honestly, I wasn't super impressed with the outputs. You want to guess how much that bill is every month?

### [Kevin Williams · Passage 36](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-36)

Well. I would guess $5,000 maybe somewhere in there.

### [Elijah Szasz · Passage 37](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-37)

Oh, so yeah, man, between five and $7,000 a month. Those are starting to be real costs, right? And he argued, hey, these would be three different employees that I'd be paying 80 to $100,000 a year, which I also kind of had a question looking at the output of the work. I don't know. Or 10,000 in the Philippines. I don't know. But yeah, yeah, that.

### [Kevin Williams · Passage 38](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-38)

Yeah, so.

### [Elijah Szasz · Passage 39](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-39)

That turns into a real line item. Now that's a 50 to $60,000 a year line item on your P &amp;L for token costs that used to be $200 a month. And then a switch got flipped. And for a small business or a business just getting out of the gates, that's very material. Like that's very material.

### [Kevin Williams · Passage 40](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-40)

would think that would be just a full-on blocker because this still isn't easy. Even ⁓ consultancies, ⁓ and trying to get people activated with this stuff is we exist, I guess. So yeah, that's going to be a big ask. But I do maintain that you're going to see a decline ⁓ good enough quality. And good enough quality is going to be ⁓ what it was just a year ago. And still have meta in the wings. So they renamed their llama series because previous llamas were sad. So now I forget exactly what they're calling it. ⁓ at the

### [Elijah Szasz · Passage 41](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-41)

Total brand refresh that I just get rid of that. Yeah. A little bit of negative connotation there.

### [Kevin Williams · Passage 42](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-42)

Pyridobrine refresh, yeah. They're doing interesting things with it internally, but that's the sort of model that will pop out there at some point and it'll be really cheap. Or Kimi, one of the Chinese models is really cheap. that's floating around out there. Like people are gonna find ways to make this make sense ⁓ ⁓ everybody needs the Ferrari. They truly don't and that actually might be somewhat of a problem with the business models that are predicating these the big frontier labs is that they're investing ⁓ in ⁓ all of capacity, it's anticipating a continued linear growth ⁓ in ⁓ the of models which is by the way still planned playing out and that people are going to pay for those models at that frontier price. ⁓ it's of like ⁓ you extended thinking and you see pro thinking in some of the models. Like for mere mortals who aren't doing like polymer research, there's really no reason to be using like ⁓ those sort settings in these models at this point because they're already quite good. So who cares if we're at Opus 6.0 and it's solving cancer, awesome. Let's devote the process resources to that. But for those of us who are doing Slack interactions and blogs and things like that, okay that your model isn't quite as brilliant if it's cost effective enough. So ⁓ optimistic, is I think that my point there that this will normalize in a way that the variable costs of operating these models as long as you're thinking about that over time. going to work out, but it will get more expensive from a fixed cost perspective in the short term.

### [Elijah Szasz · Passage 43](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-43)

Totally, totally. And I wanted to run this by you too, man. I thought this was really interesting. I just love talking to people and seeing how they're actually using AI at work, right? And then looking at the outputs of what they're getting, looking at how they wrap their heads around the cost structures and where this is going to go for them in their business. And one of the traps I constantly fall into is I will open up my social media feed. and see all the no code bros going on and on. used to be N8N, then it was like, oh, everybody's using OpenClaw and now everybody's building custom agents, so they're using cowork and these stories about, you know, I did this, this and this, and my agent did all this while I was sleeping and these were the outputs and it's running my entire services business for me. And this is my customer support agent and this one manages issues and this one does all my marketing. And I actually, got to see two end results of people's agentic workflows the world of content publishing. ⁓ tricky content publishing, not just writing blogs or posting on LinkedIn, but creating videos. ⁓ there's a lot of steps that have to happen for something like that. Right. you have to have whatever that Content source is your video rendering if there's voice which there was in this case like do you hook it up to something like 11 labs and make all that work and This was a start to finish machine two different cases one of the guys you actually know Kevin's what I'll have to to you about this later offline It's pretty funny the end result was actually publishing to social channels rich media video, right and I will also note that neither of these people came from the world of being, these are video experts or content experts, or they really have some kind of authoritative taste as to what good content is, like what's going to perform out there. And so they jumped through all these hoops of building these agents and buying all these tokens and running this whole thing. And I'm looking at the outputs and I'm like, this is just complete AI slop. ⁓

### [Kevin Williams · Passage 44](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-44)

Hahaha

### [Elijah Szasz · Passage 45](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-45)

this is awful. This is objectively awful. Like this is garbage. Like, and listen, I'm not, you know, performance marketing expert, but I knew enough that I could look at this asset and say that ⁓ this was a creative for an ⁓ or for organic social, this would do nothing. Like people would swipe ⁓ past this because it's just, it's really, really bad, right? So. I just started thinking about where we are in that whole life cycle of these real workflows that are actually replacing people and important it is to ⁓ that human in the loop that's actually ⁓ expertise in your domain of the thing that's being produced. Because in this case, both these people thought like I am circumventing the entire a burden of a marketing team, right? Like they are going from ideation to creation to publication all on their own. And look at this. just watching my Instagram channels light up. I'm like, wow, how's that performing? Well, you know, we're just getting started. There's not a lot of likes right now. A lot of followers happening. And I'm like, can tell you why that is. Like this is really, really bad, right? Because ⁓ try to skip over that step of somebody has to have expertise in this domain to know a good, bad, a good output versus a bad output, right? And like how many other areas do you see that exact same thing happening right now with people trying to rush to the end without kind of running through this process the correct way?

### [Kevin Williams · Passage 46](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-46)

Okay, so first from a marketing pro tip perspective, if you just are generating your content via AI, you're inherently getting the median type content that's just been absorbed by the Borg and like you're just regurgitating it. And it's not like the algorithms don't recognize that. Your job as a marketer, and we're all marketers in some way, or form, actually hasn't really changed in this area, ⁓ to create new value for customers such that they engage with your brand. ⁓ you can do that really generously or you can do that by just shooting a cannon at them in some way, shape or form. ⁓ you need to be a really good marketer to lever the technologies in new ways. And have to make sure that that human are staying on top of it. And the conversation that we are having right now is new material. It doesn't exist in the LLM. And you have to find ways of injecting new material like this on top of the LLM generated content that you're doing to make it novel and differentiated and have a point of view, maybe be funny, like whatever it is. Otherwise, yeah, you're losing a nickel on every transaction and trying to make it up on volume. A slot machine that's kind of expensive, right? So.

### [Elijah Szasz · Passage 47](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-47)

You're producing a really, really fast slop machine. know, it's just like plumping out slop. Like,

### [Kevin Williams · Passage 48](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-48)

It's yeah, it's a shame and then then but using the agents not for the the peer generative task, but think about the other pieces of that that that are in the puzzle. So there's information retrieval. How do we get the right topics from different places and put those in front of it could be another LLM agent that has new fresh topics and takes and then how do we as a human look at it and make sure that it's it's well formed. There's the human in the loop. Make sure it's decent and then there's The syndication part, okay. Well, you can automate the front and the back, gathering, the information, the resources, ⁓ you can automate the syndication and placement and whatever. But ⁓ you want to create something that people actually value, ⁓ you have to pay attention in there. That mean these agents can't do really interesting things in the corner of the internet, but people are just, applying them in ways that are just kind of dumb.

### [Elijah Szasz · Passage 49](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-49)

Totally. It's almost like, just because I can, I will. And that's really not the right answer. It's, it's not always going to give you the best output or in what I saw a usable output. Like I just wouldn't publish any of these assets that I saw. so yeah, I, I just, I just thought that was, I thought that was really interesting. And I caught myself also, in that internal dialogue of like, yeah, buddy, that's just, yeah, ⁓ this is good enough. can't do that. And I was like, ⁓ wait a minute. What's that word we always say? Yet, right? ⁓ when were talking about how it is very, very useful to do, you know, I've got my content research agent, I'm finding things that are trending in a lot of conversation right now, stuff that's specific to my content pillars or my ideal client profile, right? Super, super useful, helping write the content, ⁓ helping videos for maybe it's B-roll, helping do a voiceover.

### [Kevin Williams · Passage 50](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-50)

kids.

### [Elijah Szasz · Passage 51](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-51)

all that stuff really, really good, right? But it can't really staple that all together and then just publish it without human eyes on it But you can already see a world because it does work so well in those independent steps that it might be, I don't know, six months from now where it does stitch that entire thing together and you can take your hands off of the wheel, or maybe it's a year from now. But when you start to identify all of the singular tasks that it's incredible at, but you still need that human in the loop and you got to babysit it you got to make sure it's good. That's kind of a harbinger of, okay. Yeah. ⁓ At point this really will just be all automated. Like it will get there.

### [Kevin Williams · Passage 52](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-52)

So since I'm in LA. I was, I've been taking Ubers all over the place. When I can't take Waymo's, I'm sort of obsessed with Waymo's. think they're fantastic. ⁓ But they don't go everywhere. So I ended up in an Uber last night and was talking to the driver and it's LA because, and he's script writer because of course he is, because it's LA. And we had this animated conversation about the unlock as far as storytelling. And I know I talked about this a few episodes that were in this where anybody can be a storyteller ⁓ you have access to this

### [Elijah Szasz · Passage 53](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-53)

Same, yeah. I feel safer, actually.

### [Kevin Williams · Passage 54](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-54)

technology and you can do amazing things and there are a heck of a lot more people with really good stories to tell and that's you know, not just story stories but business stories and product stories and whatever then there are people that have the technology the Tactical skills the money whatever it takes to make those happen makes this a golden age for people who have really good ideas ⁓ it also lows the bar to almost zero as far as people who think that they have good ideas And it's going to create just a sea of not just slop, but there's going to be 10 times as much gold out there, but there's going to be 100 times as much slop. the scarce resource is still attention. So focusing on media in particular, people usually think of it this way, but media is actually a bit of a real estate game. That there's only so much of it that's out there, ⁓ and only so much attention that consumers actually pay. ⁓ And that's why you have just flat out battles going on between platforms to get that eye share such that they can sell advertising. Okay, fine. But you have a hundred times the content that's heading our way. will the gold float to the top? Maybe. Or we just get overwhelmed by slop? ⁓ And conventions, ⁓ as as the way that we market and the way that we get attention,

### [Elijah Szasz · Passage 55](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-55)

Yeah. Thank you.

### [Kevin Williams · Passage 56](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-56)

just start falling apart because there's so much slop and we have to create new ways of engaging with each other. We don't really know what that looks like yet.

### [Elijah Szasz · Passage 57](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-57)

Yeah, I think the other side of that coin too is that historically, if we were introduced to a piece of software that did something really incredibly like, yeah, we're on this platform Riverside right now to record a podcast. Hasn't always been like, okay, we're not talking about it. Maybe it's not so amazing. But, you know, looking back, let's call it 10 years ago, the... the number of hurdles you had to get over in order to record and publish a podcast were a lot. And these pieces of software started coming out to solve for those different hurdles, to reduce the burden, to make it a lot easier, a lot less expensive. Now, instead of a sound engineer, then going back through the episode and making sure that everything is balanced and sounds good and does not background noise, it's just a button you press. Like AI-ified this whole thing, like make it all better, right? It's like, it's cool. That's, that's awesome. And so on the content side, you know, our brains are kind of used to thinking, all right, now I've got this tool, I can make all this content. And yeah, maybe a bunch of it slop, and it's just going to flood everybody. And there's not going to be enough attention to go around for all of this content. The other side of that coin is that the same intelligence that's powering the ability to produce this insane volume of content. could also power a lot of tools to try to get that attention, right? So it's this weird cat and mouse game where the very tech that we're using to produce the volume of this stuff is also being used in other tools to try to get people's attention. And it's almost like this parallel race happening all the time, right? And I don't know who's gonna win. I don't know how it's gonna net out. I mean, we'll just see where all the chips land, but. It's super, super interesting. And that's, think, what makes this so different is that it's not just one thing. We're not just in one lane. For every problem that it solves, it might create another one, but then that problem might also be solved by a different use of the same underlying tool set. It's very strange time in that.

### [Kevin Williams · Passage 58](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-58)

Yeah. I mean, it's a sort of like a big sigh as far as how it's going to play out. I actually I think this you've probably seen this as well, since we're surrounded by AI practitioners and internally, of course, my whole team is AI enabled. And this thing is happening as far as communication, where it's like, AI generates one email, and then AI responds to the email on the other side. And everybody knows, because it's really obvious that M dashes and whatever. But, you know, it's still faster

### [Elijah Szasz · Passage 59](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-59)

Yeah. Red. Yeah, you can talk.

### [Kevin Williams · Passage 60](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-60)

to just let Claude craft the email or let Shortwave craft the email and add my thoughts into it. Hey, respond to Sonia with X, Y, and Z. And it makes it sort of overly formal. And it's not very genuine either. You're adding salutations and all of these other pieces that are kind of pro forma, but like.

### [Elijah Szasz · Passage 61](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-61)

Mm-hmm.

### [Kevin Williams · Passage 62](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-62)

I sort of wonder if we're just gonna start shooting bullets back and forth because those are the authentic pieces of the communication as opposed to all the fluff that AI has added on top of it.

### [Elijah Szasz · Passage 63](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-63)

Yeah, yeah, I almost feel like it's one of the growing pains of AI is the fluff, the slop, the stuff to sort through. It's kind of like, you know, humans have just been handed fire ⁓ we can stay warm and cook our food and ⁓ we just burned our village down to the website. Like there's all these of this technology that we almost have to get through and learn how to work with and move around that are. getting in the way as opposed to helping. We've talked about this before too, like amount of content just internally inside of one of these companies. ⁓ I think you're a great example of this because you're always building these internal tools and then you'll have these little pauses. You're like, ⁓ my gosh, like the amount of stuff this is like producing in these Slack channels and now people have to read and understand and disseminate, make decisions about like, am I creating? work that's actually moving ⁓ ball forward or am I just keeping people stuck in the cyclical pattern of trying to keep up with ⁓ these models are throwing at them at any given day, right? It's, you know, again, like sometimes it really is like, you know, one step forward, two steps back, and then you got to say, okay, this wasn't the best use of that tool. Now I'm going to try this. Like you have to experiment.

### [Kevin Williams · Passage 64](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-64)

Yeah, and I'm a little bit torn and we're both big fans of dictation. That being like a key part of my workflows is dictating everything with Whisper Flow or whatever your tool of choice is. But that's a canon that I'm dictating 30,000 words a week. that's with ⁓ of the bots that have all the click up tasks and have all of these other pieces in place.

### [Elijah Szasz · Passage 65](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-65)

Yeah.

### [Kevin Williams · Passage 66](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-66)

And I've had to add a layer to that, that it's like, okay, there's a repository of everything that's going on. And then I need to make sure that I'm reaching out to people individually so that they can sort through the noise. ⁓ Because yes, context is better. The LLMs like more information. So being able to pull it from a bunch of places, that's better. But ⁓ more in channels like Slack is, ⁓ or or wherever you might live.

### [Elijah Szasz · Passage 67](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-67)

Yeah.

### [Kevin Williams · Passage 68](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-68)

is not necessarily better. So how do you keep people focused on the right pieces of noise or not noise and be to like ⁓ get the they need without creating new layers of complexity. And that's been one of my internal goals is to make sure that ⁓ everybody has clear idea every day of ⁓ what they're to be focused on ⁓ and everybody clear line of sight to the tools that allow them to get the information. ⁓ be that client transcripts or a click up projects or email or whatever it is such that the system doesn't break down when they need to reach across the aisle and collaborate with somebody who may or may not be available based on work scheduling, geographic location, whatever it is. So that's sort of the promised land, I think, but it is a little bit of a rough adolescence as you're getting through it. I imagine, I love my people to death because they're they bear with me.

### [Elijah Szasz · Passage 69](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-69)

Yeah.

### [Kevin Williams · Passage 70](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-70)

But have like a howitzer of information that's coming out of me and my bot army every day. And I love it. I think it's great. But being on the other end of that, wow. Sorry, guys. ⁓

### [Elijah Szasz · Passage 71](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-71)

Yeah. It's like getting the email every morning from that one manager who just writes a novel and you're like, oh my God, I'm going to have to sit here and read like 17 paragraphs in this email. Now, Jim and I will summarize that for you. that is it's analogous to a lot of what ends up getting thrown at people. You're like, wow, I've got a ton of stuff I've got to now sort through to understand what do I really need to be paying attention to? And I have to share with you, like I've been working on this agentic personal assistant that does all sorts of crazy tool calls for that very purpose of trying to cut through all the noise and get me focused on what's most important for the day. That's its end goal, working to varying degrees. But yeah, that's too long of a story, almost long as a story of what I just feel like we have to touch on before we jump. you this word and then just moved right on. But ⁓ mythos. A lot of people are going to be hearing about Mythos because this is going to have real world impacts to a lot of people's to day jobs, things they're going to be hearing about at their companies. So ⁓ TLDR is ⁓ dropped, ⁓ they didn't drop a model. They built a model, ⁓ release it to the public. The most powerful model yet, it's called Mythos. This was accidentally leaked in late March when Fortune found ⁓ draft of their blog post. on an unsecured database, is a little embarrassing, but Anthropic confirmed it. Then they officially launched this thing, um, April 8th, alongside what they're calling project glasswing. And this isn't any kind of specialized security tool. This is a general purpose reasoning model that happens to be really, really good at finding cybersecurity vulnerabilities. the big reason why they didn't release it to the public. So in the past few weeks, it found thousands of these zero-day bugs in every major OS and browser, including a 27-year-old flaw in OpenBSD, which powers all sorts of things like security and routers and thought to be the most secure thing in the world. So ⁓ about companies, including big companies like Apple and Google and Microsoft, JP Morgan Chase, They've got early access to this to use it defensively because Anthropic won't release it to the public. And this really matters at work because ⁓ piece of company, ⁓ piece of software your company runs, ⁓ CRM, your payroll, your operation systems, your email, it all has vulnerabilities that are sitting inside there right now. And Mathos can find them faster than any human security team ⁓ history ever could. And so the question starts to become, well, who's gonna find them first? And how big are these vulnerabilities? And are they in your Coinbase account? Are they in your Bank of America account? Are they in your sensitive company information? So this is kind of that future stuff that we were talking about, like one day AI is gonna da-da-da-da-da, and this is gonna be a big problem. Guess what? already here. It's here, and it's like, in a weird place where some people have this kneecapped version of this insane model. Most people don't. But when will this just be generally available ⁓ when will it become a problem? So.

### [Kevin Williams · Passage 72](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-72)

Okay, so first off. zero day exploit. It's not something that people generally know what it is, but it's when been discovered ⁓ it hasn't been fixed yet. So there's this window of time between the bad guys and the good guys. The good guys have to figure out it exists. If the bad guys figure out it exists, they happily go to town and do bad things. ⁓

### [Elijah Szasz · Passage 73](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-73)

My bad. The day you found out your door was unlocked and it's been unlocked and you just didn't even know.

### [Kevin Williams · Passage 74](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-74)

It's been like for a long time and nobody even knew about it. So that's what a zero day exploit is.

### [Elijah Szasz · Passage 75](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-75)

Harder garden keys were sitting right on the coffee table.

### [Kevin Williams · Passage 76](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-76)

So the smaller the platform, the bigger the problem. So if you use some bespoke ⁓ that ⁓ I don't know, restoration of Toyota trucks or whatever the heck this crazy thing is that you've been paying for that makes sense in your business. I some businesses that do like coffee wholesaling and they're like agriculture CRMs or ERPs that deal with agriculture. And like they're tiny little platforms that are pretty expensive. And those are going to be the last ones to fix this stuff because they don't have the big dev teams. They're not getting the heads up. They certainly don't have access to the plot, to the model, to test it. So in your universe, you want to be aware of small fry vendors that you're working with because that's where vulnerabilities are likely to come from, although they can come from anywhere. The other thing, and I'm going on my third speaking event of this week here that I've been telling everybody as a result of this is you aren't ⁓ secure passwords, you need to use secure passwords. So just do not pass go. Go download one password, last pass. You can actually lean into Apple passwords. Just use Apple passwords, okay. ⁓

### [Elijah Szasz · Passage 77](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-77)

⁓ my goodness. Such an overwriter. I was going to say if you're in the Apple ecosystem, yeah, yeah, it's free.

### [Kevin Williams · Passage 78](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-78)

In any case, these are the ones that generate the annoyingly long one-off passwords. And if you think you're clever because you have Rover with one at the end and then you add an exclamation point, you're not because as soon as something gets cracked, like let's just say it's that coffee CRM and you've used that password, even if that password isn't identical, what systems like this will do is they'll extremely rapidly create all the permutations of it.

### [Elijah Szasz · Passage 79](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-79)

Every variation, yeah. ⁓ the S, we'll try a dollar sign. one, we'll try it out like it'll just, and it'll do it so fast.

### [Kevin Williams · Passage 80](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-80)

Every variation, replace all the vowels with numbers, create, yeah. Yeah. So you have to do this. Like you really have to do this. It's painful. takes.

### [Elijah Szasz · Passage 81](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-81)

You shouldn't know your passwords. Like that's, that's a sign that you're doing it right. Is that you don't know any of your password. You have to go to the locker. You have to hit copy. You got to hit paste. Yes. It's a little bit annoying. but this, you know, I mean, this is something everybody should have ⁓ ⁓ 15 years ago. That's how long I've been using password lockers, ⁓ it's, it's now reached a boiling point. Like you have to do this. You have to do this.

### [Kevin Williams · Passage 82](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-82)

You don't know the passwords. and You know, people get very frustrated with two factor authentication. This is when Google pops up and says, check your phone or whatever. you noticed that it's happening more and more often? Well, the reason it's happening more and more often is because there are more incursion attempts that are going on that are flagging things and they're happening at scale. you know, somewhere behind the scenes, I sort of picture it like that scene in Titanic when like the ship's going down and the radio operators are running around like trying to like keep things intact.

### [Elijah Szasz · Passage 83](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-83)

⁓ for sure.

### [Kevin Williams · Passage 84](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-84)

There are poor souls in network security who are just battling these demons like all day long. And we don't, the rest of us generally don't have to deal with it. two FAs are really frustrating. I get that they interrupt your workflow, but ⁓ really do work. And at least at this stage of development, it's a pretty secure way to go about things. So in anything that touches real money, ⁓ accounts, real sensitive information, ⁓ need a one-time password.

### [Elijah Szasz · Passage 85](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-85)

⁓ man. Yeah.

### [Kevin Williams · Passage 86](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-86)

and you need some sort of 2FA in place. Like, I'm not that guy who's saying, ⁓ no, know, they're gonna, the criminals are gonna rip us off, or least I wasn't that guy. ⁓ now with a tool like this, they totally are. So. ⁓

### [Elijah Szasz · Passage 87](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-87)

Yeah. Absolutely. Yeah. And just for people who don't know, like when we talk about 2FA, that's like, you have your username, you have your password, you're to log into your software. the two factor authentication is just one more layer. The three most popular ones are that it'll send you an email, right? If your email has already been hacked, that's a problem. It could send you a text message. If your SIM card has been spoofed, that's also a problem. The most robust one is probably an authenticator app. Google has a free authenticator app. And what this thing does is It's a digital representation of way back in the day, dating myself, but you still have like these little ⁓ keys, right? And have this little plastic device with an LCD screen that would always be displaying a number that would change like every 15 seconds. And that's how you'd get into your corporate bank account or into your top secret corporate CRM or whatever it was, right? This is a version of that that lives on your phone. And so just have a list of all the different things that you log into that are dynamically generating these codes that are that. one more step for you to get in. And I'll tell you, Kevin, like I get super busy and I have noticed like, you know, over the last six months, the ask for platforms that didn't require this before saying, Hey, we recommend adding this to FAA. Do you want to do it? I'm like, ⁓ skip. I'll do it later. Now I'm like, yes, yes, yes. Like loading up this laundry list of two FAA codes in my Google auth app, because this is

### [Kevin Williams · Passage 88](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-88)

Yes.

### [Elijah Szasz · Passage 89](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-89)

This is getting real. So from the companies you've been talking to, I know we're about to run out of time here, but does this feel more optimistic or scary? Like what is some of the feedback that you're hearing so far about not only the mythos ability to be able to do this, but the fact that it's been given to a very select number of companies. And me, it kind of feels like Hey, ⁓ SMBs over here. We're scared. What's going on? And we work with lot of SMBs. What is the general feedback that you're getting for something like that? Or do people understand even the scope and scale of this?

### [Kevin Williams · Passage 90](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-90)

even people that are technical are not paying attention to this. it's even just, you know, I last call was with a CTO of a very big company. You know, everybody's busy, their heads down and they do have cyber experts who are paying attention to this stuff and they're starting to ring the bells. But, you know, the basic things will really help. So ⁓ ⁓ use one time passwords.

### [Elijah Szasz · Passage 91](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-91)

Okay. Again, just us. Just me staying up at three in morning worrying.

### [Kevin Williams · Passage 92](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-92)

I mean, just think if what happened in your small business tomorrow, if somebody just cleared out all your accounts, I mean, could you survive that? Maybe the bank will honor it. Maybe they won't. Regardless, it's going to be a nightmare. You're not going to make payroll. Like, ah, horrible, right?

### [Elijah Szasz · Passage 93](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-93)

Yeah, yeah. I mean, I might be overthinking this a little bit, but even with things like these password lockers, right? There's a lot of password lockers out there. I don't know how big any of these given companies are. I don't know how robust they are. And I actually did start to lean into the whole Apple ecosystem of saving passwords just because I'm like, okay. very big company, this would be disastrous if found a vulnerability in that, they're probably gonna be paying more attention than most. so of, again, getting the companies that have access to this model that are looking at this first and deeply, and Google as well, Like Google Authenticator, that's why I'm kind of like choosing that over SMS or email verification for 2FA. These are the kinds things to start to pay attention to. as we get into these rapid iterations of models that are surpassing what I didn't think we'd have for at least another year. And I'm like, ⁓ OK, this is here now. Amazing.

### [Kevin Williams · Passage 94](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-94)

So the other related bit here is from an analog perspective, if your organization is moving money around or if you have relatives, whatever it is, ⁓ some sort of analog passphrase that guarantees it because it's trivial to deep fake, particularly voice and even video is getting better. ⁓

### [Elijah Szasz · Passage 95](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-95)

⁓ voice is so good already and video is like really coming up too. Yeah.

### [Kevin Williams · Passage 96](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-96)

And really is starting to happen in SMBs where the CFO calls up the controller or calls up the payables clerk and basically yells at them and they transfer money because it sounds like the CFO, like it has their mannerisms and whatever. ⁓ would they doubt it? ⁓ something as simple as having that passphrase ⁓ family members and for critical transactions, totally analog, but might save you a million dollars. ⁓

### [Elijah Szasz · Passage 97](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-97)

Yeah, totally. Yeah. Absolutely. mean, yeah, yeah. This is the new version of all the spam email that I've been getting for years to our accounts payable department with just an invoice, right? And saying, oh, this is past due. You've got to pay this. And the accounts payable person is like, hey, I don't remember this vendor. I'm like, it's because it's fake. Ignore that. That's a scam. It's getting next level. Kevin, that's it for us. It's been real and we'll do it again next week. We'll see you.

### [Kevin Williams · Passage 98](https://ascendlabs.ai/podcast/ep-22-when-your-entire-ai-team-calls-in-sick-at-6-am/#transcript-98)

All right, good talk.

