Are You Ready to Let the Agent Make the Call?

Kevin Williams: It can be true that there is a ton of demand for tokens that's just going to the absolute moon. And that the business model of the data centers and the neo clouds and all of this stuff that's going on behind the scenes are fundamentally flawed because of not AI stuff, but just like finance stuff as far as interest rates. Matt: Circular finance. Kevin Williams: Matt, welcome back.

Have you Matt: Yeah. Kevin Williams: woken up and looked at your portfolio today? Matt: Yeah. dude, I don't wanna be doing that today.

No, Kevin. No, some things are best left not scrutinized every single moment of this is a bad day, definitely a lot of carnage on the street this week. Kevin Williams: It's a lot of carnage, a lot of carnage. we were chatting back and forth before this about Black October, how tech stocks and stocks in general often have pullbacks in October.

So to some degree, is this the like you know, AI apocalypse that people have been talking about as far as valuations and whatnot? it's definitely everywhere. I was hauled on to our local one of our local TV stations because they needed somebody who could talk about neo clouds and infrastructure and valuation and all of this other stuff. And I had to laugh a little bit with the reporter.

I'm like, you you you realize that I'm not a financial analyst, right? Like, like I could talk about the demand and I could tell you that there's tons of demand for the tokens, but It can be true that there is a ton of demand for tokens that's just going to the absolute moon. And that the business model of the data centers and the neo clouds and all of this stuff that's going on behind the scenes are fundamentally flawed because of not AI stuff, but just like finance stuff as far as interest rates.

Well, Matt: Circular finance. Kevin Williams: there's circular finance, but there are also the interest rates. Matt: Mm-hmm. Kevin Williams: And what it takes for these data centers to be profitable, the timelines that are involved, like again, totally not our show.

But for the audience, if you're not aware, like we everything that drives AI behind the scenes are GPUs, are like graphic processing units. You and they have a much shorter life lifespan than traditional processors do. So if it takes you three or four years to build a data center. You expect to get X number of years of value out of it, but there's a lot of concern that the depreciation of the GPUs will drive those business models into unprofitability, even if there's tons of demand, even if those aging GPUs are still processing.

It actually becomes like way more of just a technical finance type question than it does an AI question. But here we are. And you know, what's gonna happen next, right? Matt: Yeah, totally.

And then Elon added some or fodder to the fire on X when he posted that he he thinks this Neo Claude stuff really is in trouble. The thing the what did you think about like the S one with Anthropic? They didn't release their demand for Q three. So I'm with you.

Demand from my perspective is through the roof. And I c I think it's under hype. There'll be so much more demand than we're all expecting. But I was like, the the talking heads on CNBC are disputing like w maybe demand's not growing as fast as we all think it is because Q three numbers weren't released.

I don't know if you have an opinion about that or if you're just like me, Kevin, and your this stuff is like, guys, there is unlimited demand. Don't even worry about demand. Kevin Williams: That's pretty much where I am. I think the most interesting thing there is that they're sitting on a hundred and thirty billion dollars in cash.

So, Matt: Yeah. Kevin Williams: like, yes, this is not great as far as infrastructure expenses if this doesn't play out for them. But it's the most efficient growth of any startup in history at the moment. And and you know, I I wanna be reasonable here.

I'm not like a total fanboy and there's no there's no scenario that I see where this is risk free. Again, because you can have all the demand in the world, but the the fundamental like finance structure of the businesses behind the scenes still might be flawed. But, but, but okay. So this is the part that that that that that gets me is I've I think we've all been expecting that at some point there's gonna have to be a turn in the economy.

It just can't keep going red hot like this forever. And I'm about to send a kid to college. I'd be really good thing for for the economy to keep rampaging forward for for my sake, right? But it will turn.

And if it turns based on supposition, based on like AI consumption, what I expect is going to happen is a picture like a demand curve that goes up up and to the right. And You start hearing the R-word recession, you start seeing days like this, and things start folding in on themselves, and companies indeed do crimp down on their budgets, and you see a decline in demand. And then that becomes like a self-fulfilling prophecy that starts the wheel like really turning. Bad things happen as far as like the economy in general.

But what happens when bad things happen in the economy? Generally. Companies react Matt: Layoffs. Kevin Williams: really strongly with layoffs, human adjustments, capital adjustments, et cetera.

What's going to happen right now when you lay off a hundred people, a thousand people, sixty thousand people? I'm looking at you, Microsoft. Where's some of that spend gonna go? It's gonna go to AI, right?

Matt: Definitely. I I so I'm I'm with you. I have a bunch of different directions we could take this, Kevin. So one of them is, you know, are the are the data like we don't know yet, regardless of there's enormous demand.

Warren Buffin always said this. We all knew the automobile was going to be successful. We knew aerospace industry was going to be successful. But looking back at history, it was actually very hard to make money in those industries.

They were very competitive. And a lot, a lot of companies went under, went bankrupt during that period. And so it's not always that easy to figure out how to make money or how to invest, even though a technology is revolutionary. So I I I think you and I agree there's very little lack of demand.

I think it's probably under hype from a demand side. And especially we'll talk a little bit later about all the agents and how much fun we're having. But Kevin, these agents are crazy good. But regardless of that.

Do they Kevin Williams: Yeah. Matt: have a profitable business model with all these other models that can get distilled and are much cheaper? And I think that also runs into trouble with the data centers too, because if you have a model that just doesn't need as much consumption, maybe that changes the the mathematics and the unit economics for the data center as well. Now, I think on the flip side, nobody wants the the game to stop and it's all the circular finance.

So they all have an interest in each other's. survival. So maybe, you know, they'll all kind of bail each other out rather than the going through that labor cycle that you talked about. But I'm still, even if there is no downturn in the market or with these stocks or this whole AI story, I do think the labor market is in big trouble, Kevin, because these agents are just made a significant step change in the last two months.

And I think we see the proof in the pudding with Microsoft laying off 60,000. Now we gotta be a little careful. They're buyouts. So it's not like they wanna lay off sixty thousand.

They're offering buyouts, so who knows how many will accept. But they only have a hundred and twenty thousand in America. to offer half of your American workforce a buyout, I think that's a data point that if you're not using these agents, you can just look at that and see what kind of impact AI can have. Kevin Williams: So people have been predicting this AI apocalypse for a long time.

It's very, very possible that this episode won't age well. But as we both know, like we are in the unenviable position of having all of our content obsolesce within like a few minutes. So it might just keep rampaging ahead and be totally air quotes fine, and we'll see. But you know, fragility in these markets does matter to to people who are running businesses, people who are Who are listening to this more than just from their retail investment perspectives?

Like bigger companies are making major, major investments in terms of infrastructure in anthropic, in open AI, whatever it might be. And if those business models end up imperiled, that could be a real problem. Now, there's a lot of transferability and switchability, at least at this point in time. I do frankly believe that it's going to get harder and harder to switch over time.

But it's still relatively easy to switch. What I'm more worried about, particularly for our listeners who tend to be more like mid-market, is you know, your vendors that are out there that you're building like, you know, you you've onboarded some cool, like, I'm not gonna say vibe coded solution, but possibly a vibe coded solution that is a SaaS product and it's run by, you know, three guys in Estonia and some duct tape. And their costs end up upside down. And now something that you've baked right into the rest of your process goes belly up.

and I do encourage everyone to be looking at their tech stack and looking at their acquisition stack and thinking, you know, what would happen if this like cool little app that everybody's come to rely on were to go belly up now that we've changed our staffing model. We've changed organizational structure about it. Do you have risks like that in your business? Matt: Super interesting.

So definitely I I think people it's very easy to make a company look legit. You have a landing page, it looks beautiful. There's a video. Everything looks like, this is a well established company.

But I think more than ever, you need to question the longevity, also the security of these small little vibe coded apps. Kevin Williams: Mm-hmm. Matt: I think is very sketchy, very prone to security hacks. So like it at for us and for a lot of the people in my circles, we're not.

buying any of these micro SAS products anymore. We're just, building it all ourselves. You mentioned specifically you think a lot of SMBs are doing the opposite. They may be taking these tools off the shelf.

But I don't know, Kevin, the agents are getting so good, I'm not sure, what you would need these little tools for. again, I know I'm in a different group and category, but are you seeing this with SMBs that you speak with? Yeah. Kevin Williams: absolutely.

Yeah. I'm a little bigger SMB or, you know, whatever, Matt: Mm-hmm. Kevin Williams: twenty million plus. Like that's not super small.

You're not talking like a florist. You're talking organizations that might have 50, 100 people that are working there. There's a lot going on in a lot of different directions. but most don't actually have the capability to build this stuff yet.

Yeah, yeah. Yes, okay. You can pop open your Matt: Mm-hmm. Yeah.

Kevin Williams: dot or you can to pop over your duet. They don't know what that is yet. okay, Matt: Mm-hmm. Kevin Williams: there are plenty of organizations that do, and plenty of organizations that are hustlers and they're like building all of their own solutions.

But most organizations are just simply not built to be constructing this stuff, even as Matt: Yeah. Kevin Williams: it becomes easy to construct it. Like, yeah, now you can have an orchestration agent that helps you design whatever it is and off it goes. But like Are you really asking the right questions?

Do you really understand your own processes? Do you really understand the outputs? Do you understand security at all? We're not in a point yet where you can where most people can do that.

Are we going to get there? I think this sort of broaches into the agent conversation a little bit to give people a sense of of Matt: P Kevin Williams: what it looks like right now. Because if you're a manager, an owner, senior leader, like what we're talking about is what sort of future you. Is going to be seeing in the next 12 months or so.

And what that means, I think, is something that you and I and everybody is going to be teasing out. Matt: And so good segue to agents. I I don't remember the last time I had a problem, searched Google for a SaaS solution. Like I re I don't know about you, but literally any problem that pops in my head, I'm just like, okay, codex, off to the races, let's solve solve this.

no, of course there's bigger software packages. What's that? Kevin Williams: In fairness, you run a development organization. So slightly different than running a big florist.

Matt: Yeah, yeah, yeah. But I think I think it's helpful for people to sec second way into agentic w the new capabilities. Again, folks, over the last two months, Agentic AI has had a really large step change in capabilities. and I've just been blown away.

I have it in Slack, I have it as a phone number, I have it on my iMessage, I have it on email. And it is incredible. I'm gonna do a little presentation actually next week, Kevin. I'm gonna show off all the crazy things, you know, simple examples of having it in a group chat with my wife and booking a massage.

Actually, this morning we woke up and we're like, hey, let's do something for Halloween. You know, agent in our group chat go off and find some stuff for us to do. Agent core hit up, you know, five different babysitters, see who's available. So there's that simple example, but then in Slack, I mean, Kevin, this is so powerful.

Like you can And you it takes I think it's still Okay. Kevin Williams: I'm gonna I'm gonna interject because it is breaking the rules of the show to say something like it does this and it does this. What is it, Matt? Matt: So the way I have it configured is hooked up to my my codex.

I'm not I have my so Kevin Williams: Okay, so it's a codec agent we're talking about here. Matt: Yeah, and I think there's a couple I probably my comp my setup's a little too complex for our listeners. There's if you're in ChatGPT, there's a thing called dot. You can add that to your Slack, and that works pretty good.

and Claude has something similar. So those are your off-the-shelf solutions. There's other tools like Victor, and not Muse is more consumer-based, but Instinct is also a good one. Those are kind of your off-the-shelf agentic solutions.

What's interesting for me is I I've set up a more complex setup where I've given it full access to my codecs with my entire server and everything logged in, authenticated credit cards. And when you when you just give it everything and you tell it like full autonomy, I mean I'm sure some bad things will happen eventually, but it has been really good. The guardrails are really tight and it's not like that open claw experience we had earlier in 2026. Kevin Williams: So so let's be a little bit specific.

So okay, you so I I'll say you can configure a dot in Slack to have very broad access without really doing too much custom configuration. that's how I'm doing a lot of things. But Matt: Mm-hmm. Kevin Williams: sort of describe how that works a little bit.

So so from a technical perspective, GPT's dot is its orchestration agent. it's very, very straightforward to get it added to a Slack channel. Not everybody here listens to uses Slack. I assume their Teams connections and I know there are Discord connections.

If you happen to be a weird business that's running itself off Discord, which does exist, you can get these agents into your individual channels. So for instance, Matt and I have a podcast like prep channel. So now take it from there. So if you add your dot or you add your agent to that channel, what does that experience look like?

How are you levering it? How are you levering that across the team? Matt: Yeah, so a couple of things. You can tag it.

So team will ask Matt, you what's the answer to this question? and I'm the one, the only one that has access to a particular system. Or maybe there's just a few people that have access, but really they need me, my my contacts. So you can tag it right in a channel.

That's extremely helpful. People often ask me to do small tasks and I can just tag it to get the task done. I can have it respond directly in the thread. I can have a share screenshots as this proof of work.

I think the difference for me is between the dot and my particular setup is dot is going to inherently have some some restrictions and guardrails. And so what people experience is they ask their dot to do something and then it's gonna ask them like 10 questions and confirmations of like, I need to access your computer. I need to go into this. Do you allow me?

Do you allow me? Do you allow me? And you can try to configure your dot to have less of those those controls, but it's it's it's kind of not. a configurable in an interface within your your chat GPT instance.

So what I did is I use a different tool that just tunnels it directly to my codex. And so we skip all the guardrails and restrictions that are on the dot and it can just go directly to the my codex and codex is, you know, super powerful and can take over my computer and access my browser and do everything I need it to do, even my credit cards. I don't know, is that too scary, Kevin? What you think?

Come on, man. Kevin Williams: that's horrifying. but but so what he's describing is he's describing a Slack app. So internally they've built a custom app, and the custom app is triggering basically an an MCP that's allowing him to create agentic functions behind the scenes.

And there are probably either skills, which we don't talk enough about, or there Matt: Tons, yeah. Kevin Williams: are Prompt structures that are built into there that strip away a lot of the guardrails. Don't do this unless you're Matt. Unless you're Matt.

It's like Matt: Bad and bad bad advice. Kevin Williams: most certainly don't hand this to like a broad team. If you're like a solopreneur and you want to like YOLO, or if you're highly technical, like Matt's team, like there are ways to put in guardrails that that like make sense. but at the very least, if you're gonna give it access to credit cards, like I know Matt's team uses ramp.

have you know, cred if you're going to put a credit card into an agent, particularly a YOLO agent like this, make sure that it has like its own virtual credit card so it can't overspend and decide that it needs to buy the Brooklyn Bill Brooklyn Bridge or whatever the heck it is. I get super annoyed with the guardrails because I want to move really, really fast. Matt: Yeah, yeah. Kevin Williams: but I st I do give them a glance.

I did have a little bit of a rage minute with moment with Dot last night as it stopped for something stupid. But like yeah. Anyway, so you get you get an agent in your chats. Matt: Well, well, I think that's the problem, Kevin, is people can't experience a gentic AI the way you and I are experiencing it because the guardrails actually just make it slower for a lot of things.

Like, go book a massage. Okay, now you're gonna ask me 15 ti questions about this. No, it's right down the road. I don't remember the name of the place.

It's like just go search it. It's the only place like within a half a mile of my house. just book with the owner, don't book with one of the employees, and here's your credit card and go. And it did it one prompt.

Bam. Done. Now, if somebody else did that, they would be frustrated because it asked them 15 questions and there's they're like, they walk away thinking AI is stupid and none of this makes any sense. I of course this stuff is overhyped and it's gonna go down to zero.

Kevin Williams: Okay, okay, this brings up a really important point. And I was just, I was just not having it out, but I was having this conversation with somebody about this yesterday about the advent of the bot sitter. So a bot sitter is somebody who is hired to be a quote, human in the loop feedback mechanism that is supposed to oversee highly complex operations that are flowing through an organization. and you see these in anywhere where there's High security, privacy issues, HIPAA issues, et cetera.

Sometimes they are like a regulatory solve in that you like you you lawyers are worried, so we're gonna like inject people. But it, but but it turns Matt: Or insurance policies or whatever. Kevin Williams: out that people are really bad bot setters. Because the problem with being a bot setter is you're you're seeing fragmented pieces of information.

So you and I were having a call yesterday about somebody, a client that we're collaborating on. You and I both have context, or actually I was able to bring you up to speed very quickly on what was going on with it. And from there, you have context on the different pieces. If instead, Matt: Mm-hmm.

Kevin Williams: in terms of an automation, you were seeing like bits and pieces of those outputs, it would be almost impossible for you to piece this stuff together. In a way that you could confidently say, thumbs up, this is a good idea, thumbs down, this is a bad idea. So this actually all lends itself to a higher level of autonomy in that if you take a deep breath and you realize that the guardrails are not that useful because you you you don't really understand them anyway, because you don't have enough context.

You're better off building an automation that understands the guardrails really well and can triage them in a way that it only surfaces things that are truly important. And then you automate more of the task. This is really, really, really bad as far as how organizations are probably going to go about lower scale automation because. It is more effective to totally remove the human from the process than it is to try and figure out how we're going to keep people in this outside of, you know, key, key human inflection points of actual client interaction or you know, a final hiring decision or or whatever it is.

But that's kind of what you're describing. I'm I'm almost convincing myself that yeah, the YOLO way is the way to go, but to build the For our older listeners, yellow means you only live once. Matt: So that that's like a mode that you can you can actually instruct your agent to just operate in YOLO. So like that essentially Kevin Williams: To in your Yeah.

Matt: means it's gonna ask you less questions and confirmations and approvals as it does its work. Kevin Williams: But I'm buying into it. What I'm not buying into is that at this stage, this is appropriate. That you can, you, you can do this.

I could probably do this. Some of our listeners that I know are are weeklies could do this. But this is as it continues to bake, like, call it six months, call it 12 months. Like these you're you're probably going to be seeing less of these decisions because there's going to be a higher level of confidence that it's not going to go off in like.

gank your credit card and do something stupid, that it's not going to book you a massage in the wrong state or a foot massage instead of a shoulder massage. Or like, you know, those those are the things that will happen if you just kind of like let it go and make decisions on its own. so there that lends itself to a question. Matt: say six months ago, yes.

l less so today. Vastly less so. Yeah. But still not Kevin Williams: Less so today and less so tomorrow, but we're still not quite there Matt: perfect.

Kevin Williams: in terms it, but it's gotta be near perfect. So, you know, you and I live and breathe this stuff, but if you go into any of our clients, they're really busy doing their stuff, right? Matt: Mm-hmm. Kevin Williams: Like having the time to worry about this, Like they they need to have a really con a high level of confidence.

Okay, okay, okay. So let's get them there. Let's say you know, hey CMO. now you can YOLO pretty much anything you want and set it free to do things.

Let's think a little bit about you to do this, you're probably a little bit farther on the curve, on the AI curve, which means you're probably doing some kind of interesting things already, as far as like mono threads and your agentic harnesses and kind of pulling things in. Just I'm asking you, I guess, Matt, like once you can truly get to that point of autonomy, what does that mean for a senior leader? What does that mean as far as Matt: Just Kevin Williams: just systems management?

Matt: Perfect. So I think, a lot of SB SMB owners are stuck in the day to day and they hear constantly from the gurus, you gotta work on the business, not in the business, right? And so this stuff really frees you up as a leader to do thinking work, right? And I really see a path where, maybe 50% of my days clicking buttons on the computer, getting various things for people or like for myself or gathering notes, whatever it is.

I spent this morning just diving deep into our our financials. It's a it's a kind of time of the month where I review all that stuff. I was able to just set up my agents. I got a bunch of stuff.

I am seeing huge productivity gains with myself personally, with my workflows that are helping me l stay working on the business, not in the business. That's how I'm experiencing it. And I've never been happier. It's crazy.

I'm s I'm Literally, Kevin, I'm just mind blown. I I've been like this for the last two months about these agents. I've and again, I know I'm an advanced user, but not and I know I'm experiencing it differently because I don't have to babysit it fifteen approvals to get one thing done. But it's really this stuff is making a huge impact for people.

Kevin Williams: So confession time, a little bit Matt: Yeah. Kevin Williams: we our our our pledges with we're going to be candid about this stuff is I have some highly sophisticated workflows. I have some highly sophisticated internal apps that I use to run our business. In fact, there are over 50 of them.

I think you have even more. We've we've sworn that we're gonna do little screen sharing at some point Matt: Yeah. Kevin Williams: so we can see under each other's hoods to see what we've built. And they're all super cool.

But I actually run most of my day through dot, which is not. Perfect. It's not as good as Grokbot in some ways. No, no, no, no, no, no, no, no, no.

But but Matt: Kevin, I gotta get you on mind because I'm I agree dot is not perfect. Kevin Williams: but this is for this is for the audience. Like I I I mean, honestly, it's fine. Matt: Well, this is the thing, this could be a prompt, we could share it with the audience.

We could make a prompt of how to duplicate this kind of setup. Kevin Williams: man. So but but even so, like, like for forget the YOLO thing. Like, I can now orchestrate almost my whole day, and I do a lot of it by by audio.

So you talked about phones. So with DOT, there's Matt: Mm-hmm. Kevin Williams: a dial-in number. It is a little weird that it's a phone number.

I haven't actually bothered to figure out why that is, but you know, I spend at least an hour of a day, the day kicked back in this this fine Liber Novo chair that that our sponsor has gifted me. to like just talk. And it's wild, like absolutely wild. Just kind of close my eyes and talk through different challenges, talk through different projects and and whatnot, fully liberating that it's Matt: Mm-hmm.

Kevin Williams: you know, I'm able to trigger different workflows. I'm able to, you know, it's it's even casual like administrative stuff. Like, you know, hey dot, send that NDA to Matt, use the last version that we had. it's somewhere in my admin folder on my Mac.

Go find it. And it is like that fast and loose. And then it'll come back with, do you want to send that to e signatures? Yeah, okay.

And yeah, I I did some work and like connecting e signatures to the NDA and you know, yada yada yada and whatever. But at this point I just talk my way through my day. And that's crazy. Matt: Yep.

And I I'm I agree. My my the the layout of my screen has completely changed. Before I'd be mostly my browser and Slack. Now I have seventeen codex windows open, because I'm just toggling between the different tasks.

Okay. This task go. Okay, now I'm onto task, go. And I just have, you know, you can people will say different agents, but it's really just I have different windows for codec, so I can keep the seventeen different parallel tasks straight in my head about what's happening.

Kevin Williams: I I needed a little bit of alignment for people who who who are not doing that approach. So if you're using a dot in codec and duet works the same in in in Claude, by the way. So duet just launched. If you don't have duet yet, you need to go find the wait list and get on the wait list and and they will eventually give you duet, but it won't necessarily happen automatically.

these are orchestrating agents. So they're serve two different ways to go. One is you use the orchestration agent. As just like a single massive thread of now, we need to do this and now we need to do that, and now we need to code this.

But what it's doing under the hood is it's opening up individual sessions. And those sessions, you could call them agents if you want, but they are doing specific things. And then you can choose if you want to go graze on a specific project, you can dive into the specific project and do refinements on it. Or you can just stay in your top level dot and tell it to do things.

I do find that that's a little fraught, that it will lose track of things over time. And me too, yeah. Matt: I I lose track of the things in my dot if I gave it seventeen different things. So that's why I separate windows for each one.

So I can keep a status in my head. Not in my head, sorry. Kevin Williams: But you'll see them in your left bar. And as of a few days ago, and and I I apologize if I somehow missed this, but this stuff changes so fast.

you know, dot is kind of cute. You have your little logo. I'll admit, mine is a small frog. It's a green frog with like green deals, and you see what's being run by dot, and it's a little icon, in my case, of a frog.

So all the way down the left bar, frog, frog, frog, frog, and then you can see the things that I'm running. This is just a point in time. Like I know this is just a point in time. Because the the use case is going to orient around that that like centralized agent that will be faster Matt: Mm-hmm.

Kevin Williams: and clearer and understand what's going on. It also chooses its own model. So you can see in your usage which model it's actually choosing. And it so far mine tends to Sol, but it will use Terra and it will use Luna, which are the lower grade OpenAI models for various things.

Which makes it more efficient for my costs, which is still highly subsidized. We haven't talked about that in a few weeks, can't forget that. but it's also more efficient for open AI, that you're not like taking a can into a pop gun type task. Matt: Mm-hmm.

Yeah, I do I do like for a lot of my tasks still having control over which model I'm using. Kevin Williams: Mm-hmm. Matt: I also love it's important for me to be able to be working in particular projects too. so it has all that particular context for what I'm working on.

So for folks who go ahead. Kevin Williams: Yeah. No, no, I know. I I was just thinking to myself that I just got a little bit fanboy about open AI.

Matt: Yeah. Kevin Williams: I actually got some hate from some not hate, but shade from some people about about that. my daily coding driver, if I'm gonna actually build something complicated, is is definitely Opus five point five is on on the cloud side. It's amazing, it's super strong, Matt: And that's what we we use as well internally.

We still use Cloud for anything code related. Kevin Williams: it's faster than it was before. It's still a little slower than codecs in some ways. And that lack of orchestration can be painful.

So just from a personal, like I shouldn't have done this before our podcast, but I kicked off a pretty big project. Just wondering why like the video stalls. Yeah, like everything freezes because Matt: This is why I don't see you on my screen. Kevin Williams: my poor CPU.

but I kicked off a pretty big project in Claude before I launched because it was going to require a design. There was going to be a UX to it. It was going to be pretty complicated in a lot of ways. And I I wanted it to get done right relatively quickly.

That for me is different from a lot of like pruning, tuning, like little stuff. Again, 50 some odd like internal apps. There's always sort of something that's going on that needs like little tweaks. And those Matt: Mm-hmm.

Kevin Williams: little tweaks, I've it may not be time efficient, it may not be token efficient. But I find codecs far easier to use as far as administering those small changes over the course of the day. it didn't have to get done in the next 10 minutes, and it Matt: Mm-hmm. Kevin Williams: just sort of dribbles out during the day.

It's not a very, very kind description of it, but it does sort of dribble out over the day that it gets stuff done on your behalf. And but that's where we're heading. So you manager, like if It w just look Leo, less you're listening to this. You're driving to work, you're, whatever.

You're thinking about your tomorrow or the things you have to do. Like, how many of those things could you do simply by dictating into one of these agents? Have you asked yourself that Matt: Yep. And I think Kevin Williams: lately?

Matt: I think that's a great point that each individual I this has been an interesting journey for me over the last two months to switch my mind from like, yeah, let me go do that. Wait, wait, wait, wait. Why am I doing this? so like it does take some behavioral changes for people individually to stop themselves from just breaking into their old habits of like, I'm just gonna go do it.

And yes, sometimes it is easier just to go do it yourself. But I think You're missing the point of this stuff is getting good really, really quickly. And it's gonna take time for you to shift your behavior and start to be able to just remember to okay, I should go, I should tag it, I should phrase it in this way. Now aga I again, like you said, Kevin, a lot of this stuff you can just say pretty loosely goosey and it'll figure it out.

But sometimes I want to phrase it in a very particular way. I want you to go this inbox, I want you to find this person in this way or this database. So I think that change f in behavior, when we first learned how to Google, there's a way to type things into Google, right? you know, and that just takes a little practice.

It's nothing that crazy, but I Kevin Williams: But I think we can get more and more fast and loose over time. it's it is I I find myself getting super lazy. Like, yeah, just go find it, you know, like just look around Matt: for sure. Yeah, yeah.

Yeah. Yeah, like just look around. I don't know, yeah. Kevin Williams: as if you had, you know, again, this this this this the world's most attentive, like executive assistant who was tenacious and knew everything Matt: Mm-hmm.

Kevin Williams: about everything and you know, go figure it out. Are your systems connected enough to allow that? So probably half the people who are listening to this are like, yeah, this is cute. But yeah, I I there's no Matt: Yeah.

Kevin Williams: way my IT department is gonna let this stuff happen. Matt: Mm-hmm. Kevin Williams: So for you the who are in that situation, like this is a bit of the art of the possible as far as where this is likely to go. But it's it Matt: Yeah, knowing what what is possible out there.

I think Kevin Williams: is happening fast. Matt: I think another super interesting thing, Kevin, is the once you have all that data in that context, when I was in a a particular Slack thread, it somehow I was very surprised because usually when I command it in a Slack thread, it's just going off the information in that thread and some, you know, general context about our business and everything it knows about me. But it actually was able to recall. a a message a message in that Slack channel from a few days ago, which was like brand new information that was relevant to what we were doing in the thread.

And I was quite surprised because generally, you know, we talk about context in very broad strokes. And so for me to be commanding it in a thread where I assume that most of its information is just related to that thread and the general context of the business, it was actually able to pull in a few key pieces of information from a day prior. that was happening in the channel in Slack. and so that was really fascinating to me that the its ability to kind of scan and internalize new information as it's coming in through calls, emails, or Slack messages is is really improving.

And then I think that's what's enabling the proactive aspect of your dot or your duet. Like, I don't know for you, but like my dot has flagged proactively a lot of really interesting things for me. I was like, wow. Like I I sh It a flight change.

I was flying to Italy last week, as I missed you. Sorry, Kevin. My flight was changed. I almost never check.

I'm just like, whatever. and it it it got moved up by an hour. And I had to change the transfer that I was on to be able to get there. So that's just a quick example of it's continuously monitoring everything all the data sources, all the new data, and then able to proactively give you some tips and tricks for the most important things on your day or what's coming up.

so I've been really impressed by that so far. Kevin Williams: So I I'm gonna bring this in as far as the art of the possible because this is on the extreme Matt: Mm-hmm. Kevin Williams: end. Admittedly, most of us, like our day-to-day activities as important as we feel that we are, where we even as like business owners, there's a lot of like shuffling of documents and things like that around that isn't rocket science.

The rocket Matt: Yep. Kevin Williams: science comes in, the strategic thoughts, like product design, like i et cetera. That's that's why we earn our keep, I would I would I would like to say. But I'm not sure if you caught that anthropic two days ago bought methril medicine.

And what methral medicine Matt: No. Kevin Williams: is, this is wild. It is it's a biotech startup that was started not by a biotechnologist, but by a Harvard MBA 18 months ago. And he developed two different therapeutics for something for MS.

and he did it as a solopreneur. Admittedly, the guy has a biotech background. He's brilliant. You know, this is not like this is not like your average Joe out there, but Matt: yeah, I was like what?

Kevin Williams: but he administered all of the the the lab flow. So he used Claude and workflows to coordinate 11 different external labs who probably had no idea that he was like a one-man show. I don't know if it was truly a one-man show, but it was not like. a lot of people.

It it's being purported Matt: Mm-hmm. Kevin Williams: that he was a one man show. He was managed he had consultants. he got all of this done in like eighteen months and it would have taken years.

Matt: Well w I'm confused. What did he do? He d invented a new drug or? Kevin Williams: He advanced a treatment for rare forms of MS.

yeah, basically designing. Yeah, it's like a pharmaceutical approach. And Anthropic, Matt: Treatments are usually like pharmaceutical products. Okay.

Kevin Williams: but this is I mean, this is another deal. Anthropic bought him for who knows how much. I'm sure it was a lot. Matt: I'm confused why they bought him.

Kevin Williams: Because they want to be able to start doing this. So he just demonstrated in a box how you could take an enormously complicated high regulatory burden type approach to like molecular design and how you can do it as somebody who is like business oriented. So basically, Anthropic is probably going to spin up its own version of a pharmaceutical company using like the approaches that this guy took. Why do we care about that?

We care about that because the stuff we're doing, as much as cool as we think we are, is nowhere even in the neighborhood of what this guy is doing. And he's like the ability to automate that level complexity. So you who are sitting in your organization thinking there's no way that we could automate all of this, it would just be way too complicated. Like, let me introduce you to my friend Cody Tren Barger.

He's a billionaire now, so it might be hard to find, but like. That's wild, right? So that's that's where Matt: Yeah, it is Kevin Williams: we are, that the art of the possible is is so far ahead of where most people are with these tools. Matt: Well, I I this brings us to a topic we had a few weeks ago about how the frontier has, generative text is solved, coding is solved, self-driving cars are solved, agentic AI is now officially solved.

and so the frontier has shifted to math, science, biology. And I I I Kevin, literally, I feel like every three days a new mathematical proof that has been, you know, mathematicians have been trying to solve for a hundred years gets solved. And there was a huge breakthrough in computer science proof just last week or two. and so that, you know, gives me further evidence that this stuff will continue to self-improve.

but I I just I think biology is about to go through a massive renaissance and we might be able to live. I know this sounds crazy, but we might be able to live until like 120, 150 years old. Like I I'm what do you think? I'm like really Kevin Williams: Yeah yeah.

I mean, hopefully we get some positive benefits out of this. I mean, you and I are we're super excited, we're running neat businesses, like we're running around solving problems, but the externalities of this stuff is real. So there's gotta Matt: Yeah. Kevin Williams: be there's gotta be a trade, and that's the utopian view is that we're going to accelerate all sciences in new and exciting ways that were just inconceivable before.

I Matt: Yep. Kevin Williams: was actually trying to describe this to one of my kids the other day that like There have been brilliant points of light throughout history. You've got the Newtons and the Galileos and the Einsteins that that that human spark of ingenuity led to new breakthroughs that led to generations of scientific progress. But there's been a lot of that over the last 500 years, and now everything has become so enormously complex that the new breakthroughs are less likely to come from intuition.

Than they are to come from mass data consumption. So Matt: Yep. Kevin Williams: given the level of complexity, we're at the right time of human scientific development, such that you can apply these models on top of them and start amplifying all of these sciences in new ways. Where humans, this is it's not fair and it's not true to say that humans would be capped out, but the Galileos and the Newtons and et cetera, are still super, Matt: Just take a while.

Kevin Williams: super rare. So if if they're both rare. and the complexity that we're dealing with is exponentially higher than what they were dealing with, those breakthroughs wouldn't happen at the pace that they would using these technologies. So that's really, Matt: Is Kevin Williams: really exciting.

Matt: and I I think specifically in biology, because with physics and math, there's a certain level of simplicity that comes with, a a type of Newton Newtonian breakthrough or relativity breakthrough. Like those are like relatively like simple, simple Kevin Williams: Newton just flopped over. Matt: concepts. Like they were brilliant, truly, truly brilliant.

But I think with biology, you're spot on. There is so much enormous amount of data. The human body is so co complex that AI Will make a massive, massive leap forward, I think, for us biologically. Salt curing disease.

I don't know if we can reverse aging, but you know, drastically slowing down aging. And yeah, this is something that I'm I'm sort of like thinking in my head, forget AI service businesses. I'm just gonna go start a longevity startup. Kevin Williams: So in that we talked about the Anthropic S one earlier.

one of the things that was in there is they spend seventy-seven percent of their compute budget on internal actions. And that's of course model training, but it is also basically applying their own technology to the rapid development of the next generation of the technology and building their own apps and features. Matt: Mm. Kevin Williams: So it's popped up in the news several times in the last six months that, you know.

Like it or not, the people running these things are a lot of them are traditional dev devs. So when they come in and they're like, we need to develop this new feature, it's gonna be called dot and it's gonna do these cool things. Okay, that's gonna take us seven months using all of these tools. And then they finish it in like 32 days because they're they're even internally, they're not yet fully accustomed to what it means to have these agents amplifying their code.

So Matt: Mm-hmm. Kevin Williams: like that that where we have this really interesting convergence. And okay, forget anthropic itself, but now think of all these other big challenges. You solve these physics challenges that involve like energy transfer and energy loss.

That applies to data centers. You can make the data centers far more efficient. Like all of these things come around or quantum computing, like is the quantum computing has been the next big thing for the last like 30 years. But maybe we're going to get to quantum computing computing as a result of these technologies.

So like super, super exciting times, even for those of us mere mortals who do mere mortal things. Matt: Yeah. Yeah. It's it's exciting time to be alive.

For our audience out there, I think the takeaway is start using agents aggressively. That's for me. I think you gotta you gotta press into that 'cause it is really life changing. Yes, definitely.

Kevin Williams: Yes. Yeah, you probably need the hundred dollar plan, by the way. and this is annoying that the a lot of these features are not available on Teams accounts. So if like duet isn't yet available on a Teams account, you have to have a personal account.

Matt: Yep. Kevin Williams: that's super frustrating. It's sort of the the way that things have been. So Matt: It's super annoying.

You just got done talking about speed, Kevin. What the heck? What's taking them so long? It's been three weeks.

Kevin Williams: Yeah. Well, it's just there's their commercial realities. So yeah, but if you don't have access to these, it might be because you're still using the $25 plan instead of at least to the hundred dollar plan. which I know, like, my gosh, I'm gonna pay four times as much.

I promise you, you're going to get four times as much value if you start leaning into these things. And if you're not, you're not Matt: Mm-hmm. Kevin Williams: doing it right. So yeah, that's my challenge to the audience is if you're not using one of these tools, get out there, start using one of these tools.

we wanna dive into more of Matt and my workflows in the future, I think, but exciting time to be alive, sir. Matt: Yeah, and I will I'll just say one last thing. If you haven't been diving in for the last three years, that's totally fine. But this is the moment.

This has been the big step change in September. Now's the time to get off the sideline and get in. So all right, guys. We'll see you next Kevin Williams: Yep.

All right. See you next week. Matt: week, Kevin.