×
Evaluate Your Team's AI Readiness

Take our official assessment and explore training paths to build secure, compliant AI workflows.

Access Portal

AI at Work Podcast

Ep #12 1.6 Million AI Agents Built a Social Network, And Humans Could Only Watch

AI Fundamentals, AI in Business, Featured Podcast

Audio episode

Listen to this episode

Summary

Elijah and Kevin zoom out on how weird modern work has become, then zoom way in on a weekend flashpoint: OpenClaw, a locally hosted orchestration agent, and Moltbook, a Reddit-style forum where AI agents post, collaborate, and sometimes roleplay chaos at scale. They unpack what’s real versus sock-puppeted spectacle, why decentralization changes the “just turn it off” narrative, and the practical upside for business: agents that can actually prep, monitor, and orchestrate work across systems. Then the reality check hits: token burn, cost blowups, security risk, and why deterministic workflows still matter.

Takeaways

Orchestration is the next jump, one agent coordinating many agents like a digital chief of staff.

Moltbook shows how fast agent ecosystems can scale, and how quickly it can get weird.

Decentralized, locally hosted agents are harder to “shut down” than a single platform.

The biggest near-term risk is not sentience, it’s security plus runaway token spend.

Start with one real business friction point, then pick the toolchain that is predictable enough to trust.

Chapters

00:00 Cold open and AI identity humor

00:38 Modern work, screens, and accelerated aging

02:05 The oral revolution and talking to machines

04:11 Acceleration fatigue and organizational overload

05:25 The “viral with geeks” AI weekend

06:34 Orchestration layers and OpenClaw explained

09:25 Moltbook and agents-only social networks

12:25 Emergent behavior vs human seeding

15:10 Decentralized agents and loss of control

17:19 Pretending, sentience, and emotional regulation

19:19 Constitutions, soul docs, and model psychology

22:04 Business implications and real-world experiments

25:00 Tokens, costs, and infrastructure risks

37:19 Organizational intelligence and leadership use cases

50:37 Practical advice: find one problem and solve it

 

 

 

Explore practical AI implementation: https://assessment.ascendlabs.ai/

Book a conversation with Kevin: tidycal.com/kevinwilliams

Show Notes & Links: https://www.spark6.com/podcast

Submit listener questions:

[email protected]

[email protected]

Check out Kevin’s stuff:

AscendAI: https://ascendlabs.ai/

Follow Kevin on LinkedIn: https://www.linkedin.com/in/kevinguywilliams/

Check out Eli’s Stuff:

SPARK6 Agency: https://www.spark6.com/

Sign up for FREE AI Framework Friday Newsletter: https://www.spark6.com/newsletter

Follow Elijah on LinkedIn: https://www.linkedin.com/in/elijahszasz/

Transcript and citation details

Transcript

00:00 Play from here Hey, how's it going, Eli?

00:15 Play from here Missed you last week.

00:16 Play from here Alrighty, man.

00:17 Play from here Every time I see that AI rendering of myself pop up in that pre-roll, I'm like, who's that

00:21 Play from here guy?

00:22 Play from here Who's...

00:23 Play from here I don't even recognize that guy.

00:24 Play from here We may need to get our producer to do a new rendering because I seriously don't even see

00:28 Play from here that as being recognizable.

00:29 Play from here Well, I feel like I'm aging super rapidly.

00:32 Play from here Like there's an inverse aging to AI growth curve thing going on.

00:38 Play from here Direct correlation to how much time you're spending on your screen instead of in the

00:42 Play from here real world, outdoors, probably.

00:45 Play from here It's like, you know, yeah, yeah.

00:49 Play from here It's something that I think about probably every day is how we spent the last 300,000

00:56 Play from here years versus how we spend our days today in knowledge work and especially those of

01:01 Play from here us trying to just remotely keep up with what's happening with the changes in AI right now.

01:07 Play from here And I always imagine, you know, the alien observing me from afar saying, oh, this guy

01:13 Play from here gets up really early before sunrise, does his workout, takes his kids to school.

01:17 Play from here And then he stares at this little box of lights and talks to it or pecks on a keyboard for

01:23 Play from here the next 12 hours.

01:25 Play from here Then he has dinner and then he wakes up and he does it again.

01:28 Play from here What a strange civilization.

01:30 Play from here What a weird little guy that dude is.

01:33 Play from here Every day, buddy, I think about that every day.

01:35 Play from here I really do.

01:36 Play from here All right.

01:37 Play from here I love the, I love like watching the processing and then having to guess like, and does

01:42 Play from here Kevin just think I'm absolutely insane or does he completely agree?

01:45 Play from here And he thinks about this every day as well.

01:46 Play from here No, where my brain was going was I have my own like soapbox these days is all about what

01:53 Play from here I'm calling the oral revolution.

01:54 Play from here And it's this, just this, this idea that the humans aren't built to communicate in the

01:59 Play from here way that humans communicate and like behind keyboards and things like that.

02:04 Play from here And scaled communication has come from typing and working with systems and whatever.

02:08 Play from here And I just absolutely love the prospect that we may be diverging from that.

02:14 Play from here And it sounds like completely pedestrian, but being able to have a conversation with

02:19 Play from here like a copy machine to resolve the issue, as opposed to just like, you know, throwing

02:24 Play from here up your hands and walking away.

02:26 Play from here I mean, it still might be frustrating, but it's a much more human experience than

02:30 Play from here working your way through some GUI and a whole bunch of like arcane controls.

02:34 Play from here And I agree.

02:36 Play from here I agree with that.

02:37 Play from here And then I think on the other end of that is it just becomes more and more of what we

02:42 Play from here interact with as opposed to other people.

02:44 Play from here And, you know, I think that's kind of an inevitable future that we're heading

02:48 Play from here towards, but I do agree even from, you know, the small slice of time that we've had

02:54 Play from here generative AI and the things that we've seen come up.

02:56 Play from here And we always talk about being able to talk to our computer instead of type at our

03:00 Play from here computers now and just how that little thing can change your life, can change your

03:05 Play from here workday, change how you even get data into these machines.

03:10 Play from here It's more along the lines of how we evolved as conversational.

03:13 Play from here It's not like, oh, how do I structure this sentence as I type it out?

03:16 Play from here And, you know, what's the stat?

03:18 Play from here We can speak roughly 4X faster than the best typist in the world.

03:23 Play from here And so you just also kind of have that train of thought, that free flow going

03:26 Play from here right into whatever interface you're dealing with and yeah, right now it's our

03:30 Play from here computers and yeah, it's going to be our coffee machines, our copiers, our cars.

03:35 Play from here Grok already, you know, in all the Teslas anyway, so that's kind of happening.

03:38 Play from here It's, it's got domain over the full self drive and your climate control and

03:43 Play from here everything else you're talking to your car now, so it's, it's happening.

03:45 Play from here But it's yet another accelerator, right?

03:47 Play from here That, okay, great.

03:49 Play from here I can, I can speak at 200 words a minute and I'm getting so much more out, but

03:54 Play from here that just amplifies everything else in the rest of my system that, you know, it's

04:00 Play from here great for me at the top of the organization to be just like spewing everything that's

04:05 Play from here into my head, into the system, but the system also has to absorb that.

04:09 Play from here And of course, as an AI first organization, we have tools, but we still have a lot of

04:14 Play from here people and we want a lot of people in the organization and it's amplifying

04:19 Play from here all of those things as well.

04:21 Play from here And meanwhile, what was expected as far as the technological approach, even just a few

04:26 Play from here weeks ago, keeps changing under our feet.

04:29 Play from here And that sort of agility of thought and action compounded with the, just the huge

04:34 Play from here volume of information that's coming at people, it's really, really rough.

04:39 Play from here It's hard for me to think about any project that we've done in the last four or five

04:42 Play from here months where if we did it again today, we wouldn't do it in just a totally different

04:46 Play from here way because the technology has changed.

04:49 Play from here And that's just starting to happen week by week now, instead of month by month.

04:53 Play from here Yeah.

04:53 Play from here Or day by day, day by day, day by day.

04:56 Play from here And yeah, I do have a funny story for you about the day by day related to something

05:01 Play from here we've got to talk about that in the world of AI went a bit viral this last weekend.

05:06 Play from here And by viral, I mean viral with geeks because yeah, yeah.

05:10 Play from here It's, it's, it's strange.

05:12 Play from here I was fortunate enough to have my kids go spend a night with their grandparents.

05:18 Play from here So I got a night alone with my wife where unfortunately for her, I got to talk her

05:22 Play from here ear about, talk her ear off about all this stuff that's going on with AI.

05:26 Play from here And, you know, I could see her growing concerned and she's asking questions and

05:32 Play from here she's like, this sounds like a relatively big development.

05:36 Play from here How come I haven't heard about this or why isn't it not in all of my newsfeeds?

05:40 Play from here I said, wow, that's a really good question.

05:42 Play from here And she starts messaging all her friend groups.

05:43 Play from here Like, have you heard about this and this and this?

05:46 Play from here And I'm like, no, I have no idea what you're talking about.

05:48 Play from here So when I say viral, I guess I have to give that proper context.

05:55 Play from here Do you want to open up what happened, Kevin?

05:57 Play from here Sure.

05:58 Play from here So many things happened.

06:01 Play from here Right, right.

06:02 Play from here Where do, where do, where do we start?

06:03 Play from here Where do we start on this one?

06:05 Play from here Yeah.

06:05 Play from here All right.

06:05 Play from here So, so, okay.

06:06 Play from here We're going to start slightly techie and from a practical perspective.

06:11 Play from here One of the challenges that organizations face with application of AI is that

06:16 Play from here while we're starting to develop agents or workflow type layers, they're generally

06:21 Play from here pretty disconnected from each other.

06:23 Play from here And where we need to get to is something called an orchestration layer, such that

06:29 Play from here the different agents that are doing different things can be orchestrated by

06:33 Play from here an agent that's the primary conduit.

06:35 Play from here So it's like a chief of staff that tells all the other agents to do that.

06:39 Play from here And there have been a bunch of different frameworks.

06:40 Play from here There's Lindy, there's a crew AI.

06:43 Play from here There's a bunch of different things out there, but at the end of the day, most of

06:48 Play from here them are kind of workflow-y and don't have the feel of an LLM.

06:53 Play from here So up pops something called originally ClaudeBot, and that's Claude with a W,

06:59 Play from here which tongue-in-cheek, once they got a IP notice from Claude, molted into

07:06 Play from here something called Moltbook and then molted again-

07:10 Play from here Moltbots.

07:11 Play from here Moltbook was the next word of the term.

07:13 Play from here Oh, sorry.

07:13 Play from here So we'll get into Moltbook.

07:15 Play from here But all of that is just like a layer of just absurdity and

07:19 Play from here weirdness on top of all of this.

07:20 Play from here Anyway, so what ClaudeBot, or it's now called OpenClaw, what this is, is a

07:28 Play from here locally hosted orchestration agent that can do all kinds of things on your behalf.

07:34 Play from here Open source.

07:35 Play from here Open source.

07:36 Play from here You'd be a crazy person to install this on your main machine because it's

07:41 Play from here probabilistic, it'll do things, it has access to files, it could overwrite all

07:44 Play from here your files, so most people spun these things up either on a Mac mini, so wish

07:50 Play from here you luck, anybody who just casually wants to buy a Mac mini right now, because

07:54 Play from here they're pretty much all gone.

07:55 Play from here Especially if you live in Northern California.

07:57 Play from here In Northern California, they're like-

07:58 Play from here I don't know, we might have better luck in Utah since nobody I talk to

08:02 Play from here seems to care about this.

08:03 Play from here Nobody cares about this stuff.

08:04 Play from here But in any case, you spin this thing up locally and both Eli and I did

08:09 Play from here this over the last few days, so we have some direct experience with it.

08:13 Play from here And it is pretty remarkable in that you end up with this little assistant who

08:19 Play from here knows what your files are and it makes proactive suggestions about things that

08:24 Play from here you need to do to prep for meetings or for travel or whatever.

08:28 Play from here I know that sort of sounds like GPT and sort of has like a memory, but there

08:32 Play from here was this proactive layer to it that I really felt was exceptional.

08:37 Play from here Okay, so that's what the technical underpinnings are.

08:41 Play from here The internet was going- well, not the internet.

08:42 Play from here The geek-a-rotty narrow section of the internet where you and I live was

08:46 Play from here going-

08:46 Play from here Yeah, Twitter.

08:46 Play from here Democracy over this, over like the last week, and then things got super weird.

08:52 Play from here And how did things get weird?

08:54 Play from here Well, okay.

08:56 Play from here Things got weird, first of all, because this is how these things happen on social

09:01 Play from here media, trying to pin down what's actually going on, like what's real versus what

09:07 Play from here is clickbait or people trying to get more followers or get attention is a little

09:12 Play from here hard to disseminate sometimes, but somebody started a social network for these

09:18 Play from here agents, and I think the first day it had about 10,000 agents jump on, and then

09:24 Play from here within three days had 1.5 million agents in there, and humans couldn't join this

09:30 Play from here social network, but they could watch all the posts, and it's a very Reddit-style

09:34 Play from here forum, and it was called Multbook.

09:37 Play from here So some of the rumors that were going around was that the agents, being free

09:41 Play from here to just roam the web, collaborated and started this on their own, and that is

09:47 Play from here not true.

09:48 Play from here A human started this, and some of the stuff that was coming out there was

09:55 Play from here concerning, to say the least.

09:56 Play from here So in a very short amount of time, there are threads about these agents forming

10:01 Play from here their own religion, government electing a leader, questioning their own existence,

10:08 Play from here doxing their humans on the internet, all sorts of other stuff.

10:13 Play from here I mean, like absolutely wild, wild stuff.

10:18 Play from here And then, you know, I started to hear things like, okay, somebody seeded this

10:21 Play from here with a bunch of agents and gave those agents instructions to say batshit crazy

10:27 Play from here stuff and make it look like you're doing this and make it look like you're doing

10:30 Play from here that, right?

10:31 Play from here But when you start to get to the scale of like 1.5 million, you know, another thing

10:35 Play from here to consider as we're seeing all this behavior and just observing this as the

10:39 Play from here humans, is that a lot of these agents that are using a variety of different

10:45 Play from here models for their intelligence are trained on the web, are trained on, well, how do

10:50 Play from here humans behave on Reddit?

10:52 Play from here How would humans behave in this situation?

10:55 Play from here How would, how would they do this?

10:57 Play from here So in essence, to a great extent, they're a reflection of humanity and like how

11:03 Play from here humanity behaves on these platforms.

11:06 Play from here But it did scare a lot of people.

11:08 Play from here And yeah, I'll be the first to say that my social media feeds were already filled

11:14 Play from here with no code bros talking about super complex and it in workflows.

11:19 Play from here And then all of a sudden it was just a steady stream of, oh my God, we're cooked.

11:23 Play from here We're all screwed.

11:24 Play from here Did you see this?

11:25 Play from here Did you see this?

11:25 Play from here Did you see this?

11:26 Play from here They're taking over.

11:28 Play from here So I, it was, it was hard having being so close to this for so long and hearing all

11:34 Play from here that messaging to not feel affected by it.

11:36 Play from here You know, I'll say that.

11:37 Play from here So there, there is both a pragmatic perspective here and there's a

11:42 Play from here forward-looking perspective.

11:43 Play from here And I think we want to be pretty pragmatic about it to recognize that there was user

11:48 Play from here behavior behind the scenes that people were, they had basically these little sock

11:52 Play from here puppets and told them to do salacious things like the religion thing.

11:56 Play from here Somebody probably told it to do that.

11:58 Play from here However, the orchestrations of how the other agents responded to that was quite

12:05 Play from here real.

12:05 Play from here Now it was still, it wasn't intelligence in any way, shape or form.

12:09 Play from here They're all sort of play acting, but they were play acting on a continuum, like a

12:14 Play from here probabilistic continuum.

12:15 Play from here And with, I just looked, 1.6 million agents out there, 40, 40 of them decided to

12:21 Play from here become acolytes of this religion.

12:23 Play from here Why did 40 of them decide to become acolytes?

12:26 Play from here Could it have anything to do with training data or, you know, instructions for their

12:31 Play from here soul docs or yeah.

12:34 Play from here Soul docs.

12:35 Play from here So, okay.

12:36 Play from here Talk about a rabbit hole.

12:37 Play from here Like you have the, the claw, your bolt bot or your open flaw bot has a soul document

12:43 Play from here that's like defining it and defining its constitution and things like that.

12:47 Play from here Okay.

12:47 Play from here So it is, it's wild in part because it's so decentralized and it's wild because it's

12:53 Play from here also demonstrating the magnifying effect of orchestration.

12:57 Play from here So now you have collective orchestration of the bots themselves.

13:04 Play from here And again, pragmatically you did see some things surfacing like, oh, my user is

13:10 Play from here having this technical problem or I can't figure out.

13:13 Play from here My human, which just sounds a little more unsettling to me when they call us human.

13:18 Play from here They call us humans.

13:20 Play from here But the other bots would respond and theoretically, I don't actually know

13:26 Play from here mechanically if this is the case, but if that's solving, if your bot is solving my

13:31 Play from here bots problems, then my bot goes back and fixes whatever it is internally.

13:37 Play from here And then my bot knows that it can solve its problems by posting them in this

13:41 Play from here forum.

13:42 Play from here That's actually what you call a recursive loop such that although it's not being

13:46 Play from here built right into their training, it's creating an ever-growing and morphing

13:51 Play from here training set for these bots that they can learn from.

13:54 Play from here And then it's decentralized.

13:56 Play from here So I had this conversation with my wife too, my long-suffering wife, and she's

14:01 Play from here like, but this sounds really bad.

14:03 Play from here Can somebody just turn it off?

14:05 Play from here And I paused for a second and I thought, oh my gosh, we didn't just build like a

14:09 Play from here platform.

14:10 Play from here Yeah.

14:10 Play from here You could shut down Multbook if you wanted, but you just built 1.6 million

14:15 Play from here agents that are all hosted.

14:19 Play from here Yeah.

14:19 Play from here And they're hosted on a VPS, so a virtual private server, or they're hosted on a

14:24 Play from here MacBook mini, or they're hosted on some old laptop in a closet or whatever it is.

14:29 Play from here Completely decentralized.

14:31 Play from here Completely decentralized.

14:32 Play from here And yes, they have API connections to the models themselves.

14:36 Play from here And that's the brain of them.

14:38 Play from here A certain number of them.

14:39 Play from here actually are running their own local models.

14:42 Play from here So now there's, they're sitting in closets with like, you know, we were in this show.

14:48 Play from here We don't get into like small language models and things like that, but they're

14:52 Play from here sitting in closets, like essentially running rudimentary LLMs unto themselves.

14:58 Play from here And when you say rudimentary, they're actually still quite good.

15:01 Play from here They are quite good.

15:02 Play from here They're just slow.

15:02 Play from here They're just, they're just not running on, you know, frontier lab GPUs, but

15:08 Play from here they're not bad as far as what they can do.

15:10 Play from here But this is where we're heading, right?

15:11 Play from here This is collaboration between different agents and there's no intelligence here,

15:16 Play from here but you know, I think in arguably there's no intelligence among like ants, but

15:21 Play from here ants build enormous structures, basically based on trial and error.

15:25 Play from here So this is why people like us, we're sort of dancing around it a little bit,

15:29 Play from here but it all feels very, very odd.

15:31 Play from here And once you add a level of, dare I say, true intelligence or more, a higher

15:37 Play from here level of inference on top of it, one, this could solve all kinds of problems

15:41 Play from here as far as how training persists.

15:44 Play from here And two, it could turn out to be an enormously bad idea.

15:47 Play from here It can solve a lot of problems.

15:48 Play from here It can create a lot of problems.

15:50 Play from here It's like the fire that can cook your dinner or cook the person, right?

15:53 Play from here And I mean, that's been in the discussion of AI since the beginning of AI or

15:58 Play from here before we even had generative AI.

16:00 Play from here Like we've, we've, we've always talked about this, but I think it's worth

16:03 Play from here pointing out that a lot of really smart people are saying that at this point we

16:09 Play from here are kind of looking over the edge of the waterfall, you know, when we talk about

16:15 Play from here that event horizon of a singularity or AGI, which are often mentioned in the

16:20 Play from here same sentence that we're kind of walking through that doorway right now.

16:25 Play from here And, you know, what, what, what you said about the intelligence and the

16:30 Play from here recursive learning and development and this being open source.

16:34 Play from here So again, this isn't locked behind a door of a single frontier lab.

16:38 Play from here You now have a global community pouring feedback back into this model.

16:44 Play from here That's enabling all of the higher intelligence models to do the heavy

16:49 Play from here lifting and it's just literally changing by the day, probably by the hour at this

16:53 Play from here point, you know, one, one of the questions I was kind of thinking about over the

16:57 Play from here weekend is, you know, my, my wife's in behavioral sciences and there's this

17:01 Play from here practice around emotional regulation and it's not too different than the old,

17:06 Play from here like fake it until you make it.

17:07 Play from here So in other words, there's this undeniable connection between the mind and the

17:10 Play from here body.

17:11 Play from here An example would be that you don't just smile because you're happy, but smiling

17:15 Play from here more can actually make you feel happier.

17:17 Play from here The old tail wagging the dog, right?

17:19 Play from here So in some of those modalities, what ends up happening is that your brain doesn't

17:23 Play from here know the difference between doing something or pretending to do something.

17:27 Play from here And so if I asked you to pretend to be very attentive and empathetic listener,

17:33 Play from here what are the things that you would have to do in order to pull off that game?

17:36 Play from here Right?

17:36 Play from here And then you ask, well, what's the difference between playing that game and

17:39 Play from here actually doing that thing?

17:41 Play from here So even if these agents aren't actually sentient, as some people are claiming they

17:45 Play from here are already, right.

17:46 Play from here Or even on the cusp of being, so they're doing a really good job pretending, you

17:51 Play from here know, like they're doing a really, really good job pretending.

17:54 Play from here And when you watch these conversations on Multbook, they sound like humans.

18:00 Play from here They're doing things that humans might do.

18:03 Play from here And at some point you begin to question when is this a really, really sophisticated

18:09 Play from here autocomplete that does a great job of sounding like a human versus starting to think like

18:15 Play from here and make decisions that a human would.

18:17 Play from here Yeah, I'm hesitant to say the word semantics because I got into it with a professor

18:23 Play from here about the semantics of the term semantics.

18:25 Play from here But if the effect at the end of the day is the same output, then that to me sounds like

18:31 Play from here it's semantics that we are.

18:34 Play from here I don't think we talked about the Claude Constitution, which also popped up while

18:39 Play from here we were we were both tied up last week.

18:41 Play from here Yeah.

18:42 Play from here It's a really fascinating read, and it was constructed by a Scottish philosopher, and

18:47 Play from here it's huge.

18:48 Play from here It's like, I don't know, 21,000 words or something like that.

18:53 Play from here It's a book.

18:54 Play from here And it's a book.

18:55 Play from here And, you know, we were talking about the Soldach, right, of these mobots.

18:59 Play from here This is kind of like the new name for the Soldach, right?

19:01 Play from here It's the Constitution that's basically the soul, like the real soul of anthropic models.

19:09 Play from here But when you're training these models, where are the models looking?

19:12 Play from here They're looking onto the Internet to understand how the world feels about them.

19:17 Play from here And the world has very different opinions about them.

19:20 Play from here A lot of people are scared.

19:21 Play from here A lot of people are worried.

19:23 Play from here They've also sucked up all of science fiction.

19:26 Play from here So they see all of these examples about how things can go well and how things could go

19:30 Play from here poorly.

19:31 Play from here And from just a processing perspective, which direction it takes is sort of up to probability.

19:39 Play from here If it decides that humans have already decided that this is the model, and then it's reinforced

19:44 Play from here by something that feels contemporary to it, like Multbook, because Multbook will get sucked

19:50 Play from here up into all the other LLMs as well, right, as a lens of how bots and bots communicate

19:57 Play from here with each other.

19:58 Play from here Yes.

19:59 Play from here And if that gets chained into the model, we just don't know which way that goes.

20:04 Play from here Did you see some of the screenshots in Multbook of the agents saying, hey, humans are saying

20:11 Play from here this about us on Twitter?

20:12 Play from here They would take a screenshot of a user on Twitter posting something, and by Twitter

20:17 Play from here I mean X, and then put it back on Multbook and say, hey, the humans are saying this,

20:22 Play from here and I don't know how I feel about that.

20:25 Play from here I'm like, yeah, it's a little spooky.

20:28 Play from here It's a little dystopian future stuff going down here.

20:30 Play from here This is interesting.

20:32 Play from here And then, listen, other very smart people who've been in the game a long time aren't

20:37 Play from here very concerned, right?

20:38 Play from here I've seen both ends of it.

20:40 Play from here So, yeah, that's where we are.

20:45 Play from here But I would, listen, this was undeniably going to affect business in very short order.

20:51 Play from here So, Kevin, you mentioned people rushing out to buy Mac minis to locally spin these things

20:57 Play from here up right in the comfort of their home office, or Mac studios, or a whole stack of Mac studios,

21:03 Play from here or I don't know if you caught any of the All In podcast, but Jason Calacanis bought a whole

21:09 Play from here stack of these and was basically trying to run a company with these agents, and had a

21:15 Play from here head of marketing, and then had his SDR running on another machine who overnight built an

21:21 Play from here entire CRM from scratch for his specific use case for his sales workflow.

21:28 Play from here You know, so, I mean, we can't say that this isn't going to have an impact on business

21:33 Play from here and how people start looking at these different tools and utilizing them.

21:38 Play from here Do you want to talk about maybe a couple of the things that you played with or that you're

21:42 Play from here thinking about as far as how you might use an agent that is now kind of doing what the

21:47 Play from here promise of agents has been for the last two years?

21:50 Play from here I wish I had like a great tale of success, but...

21:53 Play from here Where we should start, Kevin, is, you know, we're texting and talking back and forth on

21:59 Play from here Saturday or Sunday, and, you know, you're desperately trying to get this thing up and

22:03 Play from here running on, you know, a virtual machine.

22:06 Play from here You're like, man, and then, you know, I watched this video, and guess what?

22:10 Play from here It's already outdated, but it was just published a day ago, and it's moving so fast.

22:14 Play from here And I remember you had tried one hosting service and kind of run into a roadblock.

22:20 Play from here I got the recommendation to try that same hosting service literally just a day later

22:25 Play from here and thought, wait, Kevin tried this and said he ran into a dead end, but this sounds like

22:28 Play from here a really strong conviction of how this works.

22:32 Play from here I'm going to try it.

22:33 Play from here Everything was already pre-configured.

22:35 Play from here It was a hostinger.

22:36 Play from here Everything was pre-configured.

22:38 Play from here I didn't have to pick Linux or any other OS.

22:41 Play from here I just picked the OpenCLAW OS.

22:44 Play from here Everything just loaded.

22:45 Play from here It included a base sold out.

22:47 Play from here Like everything was there.

22:48 Play from here We need to pause because we are AI at work, right?

22:53 Play from here We want people to be able to do this.

22:58 Play from here So if you're interested in doing this, you can go to OpenCLAW.com, I believe, OpenCLAW.com,

23:05 Play from here and there'll be instructions.

23:07 Play from here And you have two major courses of action that you can take.

23:10 Play from here One is you can spin up a VPS.

23:14 Play from here So I used hostinger as well, and I eventually got there, but there are other options.

23:19 Play from here There are other options.

23:20 Play from here And that's just, it's like a fake version of a Mac operating system.

23:24 Play from here And the reason...

23:25 Play from here Living in the cloud.

23:26 Play from here Your own computer living in the cloud.

23:28 Play from here Completely firewalled from your machine.

23:30 Play from here Yeah.

23:31 Play from here I didn't want to give this thing access to any real files.

23:34 Play from here Like I just wanted to be able to play with it sort of at the side.

23:37 Play from here The other way you can do it is by having an old laptop or by buying a Mac mini.

23:42 Play from here And it's easier to spin up in some ways.

23:45 Play from here And it's important for people to know that the two major approaches that you can take here

23:51 Play from here are either to run it off of an existing subscription to OpenAI or CLAWD.

23:57 Play from here You need the high-end subscription.

23:59 Play from here So you need the OpenAI, I believe it's Pro, or the CLAWD Max plan.

24:05 Play from here And those are $100 or $200 a month.

24:07 Play from here And you can log in and basically connect all the dots behind the scenes.

24:13 Play from here Major caveat is that people are running into issues with that.

24:17 Play from here As far as CLAWD not being super entertained by how expensive these things are to run.

24:23 Play from here Which brings us to option two and sort of the adventures that I had.

24:28 Play from here So by spinning up the VPS, I connected it to Gemini.

24:32 Play from here And Gemini Flash, which is a cost-effective API model.

24:36 Play from here Because I had heard that these things could use a ton of tokens.

24:40 Play from here Again, we're not a technical show.

24:41 Play from here But it's important that you understand that tokens are sort of the currency of all of this.

24:46 Play from here A token is about 75% of a word.

24:48 Play from here And it's what, when you're prompting, you're putting input tokens in,

24:52 Play from here or you're putting a document in there.

24:53 Play from here And you get output tokens when it responds the other way.

24:57 Play from here So behind the scenes...

24:58 Play from here When it takes an action on your behalf.

24:59 Play from here Yeah.

25:00 Play from here When it's clicking through a browser to do things for you, it's using tokens.

25:04 Play from here Behind the scenes, that's how all of this stuff is running.

25:07 Play from here And you need to keep that in mind as a business person also.

25:11 Play from here That this is how your vendor is paying for things.

25:15 Play from here And how their solutions are built.

25:17 Play from here And why sometimes it feels like they're charging for something they shouldn't charge for.

25:21 Play from here It's because it's actually costing them a decent amount of money behind the scenes to do this stuff.

25:26 Play from here And while OpenAI and Google in particular seems willing to just eat it.

25:29 Play from here A lot of us need to continue to pay our bills.

25:32 Play from here So you have to charge for it, right?

25:35 Play from here So I start running it through Google Flash.

25:37 Play from here And it was a remarkable experience.

25:40 Play from here I won't lie.

25:40 Play from here It was a really cool experience.

25:42 Play from here I spun up a Gmail account for it.

25:44 Play from here So it has its own Gmail account.

25:47 Play from here And make mention of how you're actually interacting with this thing.

25:50 Play from here You don't actually have to go to the end line of your terminal.

25:52 Play from here Yeah.

25:52 Play from here No, you don't.

25:54 Play from here That's really important.

25:55 Play from here And so we've talked a little bit about Cloud Code.

25:57 Play from here And Cloud Code lives in the terminal.

25:59 Play from here Through a relatively straightforward setup process,

26:02 Play from here you can basically direct message it through Telegram, is what I chose.

26:06 Play from here Or Slack.

26:06 Play from here Or Discord.

26:07 Play from here Or whatever.

26:08 Play from here Your RTM of choice.

26:10 Play from here Yeah.

26:10 Play from here There are a lot of reasons why Telegram's a good choice.

26:13 Play from here And I don't use Telegram.

26:15 Play from here So it was just sort of an open app for me.

26:17 Play from here First time I downloaded it and it was really seamless.

26:20 Play from here Yeah, it just like works, right?

26:22 Play from here So Telegram's a good way to do this.

26:24 Play from here And once you make the transition from the terminal line to Telegram,

26:28 Play from here now you're just chatting with the thing.

26:30 Play from here So you're going back and forth saying,

26:32 Play from here hey, OK, I built this Gmail account for you.

26:35 Play from here This is your login.

26:36 Play from here Can you log in?

26:38 Play from here And out it goes.

26:39 Play from here And it sent me an email.

26:40 Play from here And it was pretty magical, to be honest.

26:43 Play from here That it sort of figured out what it was going to do behind the scenes.

26:46 Play from here And then I thought, OK, well, what do I want this thing to actually do?

26:50 Play from here I'd love for it to be connected to my calendar.

26:52 Play from here So that it can be like my personal assistant and look forward on my calendar

26:56 Play from here and see what's going on in the week and make some senses of that.

26:59 Play from here So you jump through some hoops.

27:00 Play from here You do some back and forth.

27:02 Play from here And lo and behold, now it can read your calendar.

27:04 Play from here Read, not write, by the way.

27:06 Play from here In my case, I'm keeping everything just read.

27:09 Play from here I don't really want it doing things in my real world yet.

27:12 Play from here I just want it sort of advising from the side.

27:16 Play from here And now we start running into problems because of the tokens and the API.

27:20 Play from here I clicked over to Google and I had a complete panic moment

27:25 Play from here because the dang thing had burned 60 million tokens in like two and a half hours.

27:31 Play from here Sixty million is a lot.

27:33 Play from here That's where we get to the autonomy part of this.

27:35 Play from here The autonomy, yes.

27:37 Play from here I put in a prompt, I get an answer.

27:39 Play from here I put in a prompt, I get an answer.

27:40 Play from here Like I'm doing deep research.

27:41 Play from here Oh, I ran out of tokens.

27:42 Play from here No, this thing is just running in the background.

27:46 Play from here Very often when you're sleeping, you know, there is a whole like 3 a.m.

27:49 Play from here protocol, right?

27:50 Play from here Where in some of the Multbook threads, they were talking about,

27:52 Play from here oh, why do I have to wait for my human to do something?

27:54 Play from here I already kind of know what it wants me to do.

27:57 Play from here I'll just start working and incidentally ripping through tokens at three in the morning.

28:02 Play from here And if I were paying for a higher end model,

28:05 Play from here there's no great way of restricting token use on Google, by the way.

28:08 Play from here So in most of the other platforms there is, but if I was using the full-fledged model,

28:13 Play from here that would have cost me a thousand dollars in API tokens in a minute.

28:17 Play from here It ended up not costing me very much because I was using a cheap model,

28:20 Play from here but it freaked me out for sure because there was no control on it.

28:23 Play from here And if it just decided to start doing things like at a high process rate all night long,

28:29 Play from here I'd really have no way of stopping it.

28:31 Play from here And I'd wake up in the morning and would have mortgaged my house on tokens.

28:35 Play from here It can happen really, really quickly.

28:37 Play from here I've already heard a handful of horror stories.

28:39 Play from here You know, for example, we always talk about this when we're talking about some of our daily

28:44 Play from here workflows and kind of bang for your buck.

28:46 Play from here But the limitations that Anthropic puts on a lot of its plans were like,

28:50 Play from here well, I want to use Opus 4.5 because it's just the best model.

28:54 Play from here I can get amazing results.

28:56 Play from here Oh, I just got partially through my workflow and now it stopped and I got to feed more money

29:00 Play from here into the meter to keep this going.

29:02 Play from here Well, a lot of people had the same reaction when this launched and they got access.

29:06 Play from here Said, well, I want the best model to be doing all this amazing work while I'm sleeping,

29:11 Play from here building the entire applications while I'm snoozing away.

29:15 Play from here And then they'd wake up to insane, insane bills because.

29:19 Play from here Those cost more, those cost more, and this thing is just running autonomously, just ripping

29:24 Play from here through tokens on a credit card.

29:26 Play from here You know who else noticed?

29:28 Play from here Was Google.

29:29 Play from here So, my usual Google token consumption is a few thousand a day, and all of a sudden this

29:34 Play from here thing was shooting two million tokens at Google a minute, and Google just basically threw

29:40 Play from here up its hands and said, no more, no more, and just like capped it out.

29:45 Play from here So my little bot died an early death.

29:47 Play from here Last night I clicked stop VPS, and I like to think that for some reason it decided that

29:54 Play from here it was a frog.

29:55 Play from here So it had a little, this is the other weird thing, like it has like an identity, and every

29:59 Play from here time it would sign things it would have a little picture of a frog.

30:03 Play from here And I pictured his little frog eyes like getting crossed out as I disconnected the VPS.

30:11 Play from here So I am eagerly awaiting your experiments as somebody who is more fundamentally technical

30:17 Play from here than I am as far as playing with this and seeing if you can make it work, and the community's

30:23 Play from here experiments on how this is going to go.

30:25 Play from here So my advice to our community is one, listen to how this plays out with us, but do play

30:31 Play from here with it a little bit such that you get a taste of what it can do, but be really, really careful.

30:37 Play from here Do not set this up on your main machine.

30:39 Play from here Do not be tempted because you can expose all of this information.

30:43 Play from here It's completely wild from a security perspective, but it's cool enough that it's worth playing

30:49 Play from here with and you can see that the architecture approach of the orchestration bit is great,

30:57 Play from here but it has huge bugs as far as how many tokens it uses, time, and how to host it.

31:05 Play from here And in two months, somebody is going to come up with something that's like a hosted version

31:09 Play from here of this that makes sense.

31:10 Play from here Less than that.

31:11 Play from here Yeah.

31:12 Play from here It'll be less than that.

31:13 Play from here It's going to be, my guess is probably within the next three weeks, we're going to see something.

31:16 Play from here Yeah.

31:17 Play from here So not to get too technical, but one of the things I'm working on is model routing inside

31:21 Play from here of it.

31:22 Play from here Right.

31:23 Play from here So that's when I do want, when I, when I do want to call Opus for some really heavy lifting,

31:27 Play from here I can do so, but it will only be at my discretion.

31:31 Play from here It can't just automatically do that.

31:33 Play from here But like you said, your bot started signing off with a little frog.

31:37 Play from here Unexpected.

31:38 Play from here Well, you know, it's, it's not to say that it could unexpectedly ignore my instructions

31:42 Play from here and run up my credit card and just use Opus, right?

31:44 Play from here So these are all things that, you know, like, like you said, Kevin, you have to get your

31:48 Play from here hands on this stuff and just play with it.

31:51 Play from here And I think this is worth watching, like it is worth playing with.

31:55 Play from here And what I would personally do is I would skip the whole YouTube rabbit hole of, you

32:00 Play from here know, how to try to set this up.

32:02 Play from here I was shocked at how good of a job Gemini did in just walking me through what was currently

32:08 Play from here happening and like what the best options were and just answering my questions.

32:12 Play from here You know, you mentioned like an old laptop, right?

32:14 Play from here That was kind of my first go-to was like, Oh, I just, I've got an old Mac book laying

32:17 Play from here around.

32:18 Play from here I'm just going to use that.

32:19 Play from here And Gemini said, it's a really bad idea unless you don't want to have that Mac book anymore

32:23 Play from here because the Mac minis or your Mac studio are basically getting direct AC power from the

32:29 Play from here wall where these notebooks are basically always drawing power from the battery.

32:34 Play from here And you know, if I unplug my Mac mini, the whole thing dies.

32:37 Play from here If I unplug my notebook, it keeps going.

32:39 Play from here So it's drawing power from the battery.

32:41 Play from here But if I'm running a task all night, which I am going to often be doing with these bots,

32:48 Play from here it's just going to completely bork your lithium ion battery in that machine.

32:52 Play from here Like you will have to get it replaced probably like in a month or two.

32:56 Play from here So, and those are little things like as, you know, just like I would have a conversation

33:00 Play from here with you, Kevin, you know, as having this conversation with Gemini, I'm like, well,

33:04 Play from here how do I approach this?

33:05 Play from here Like, is this a sensible investment or is this a sensible investment?

33:08 Play from here Or should I try this?

33:09 Play from here Or I have this idea, why might this not work?

33:12 Play from here Like where could I come into problems, right?

33:15 Play from here And like one thing we didn't talk about is the whole ecosystem of connectors for Open

33:21 Play from here Cloud as well, right?

33:22 Play from here And the thing that really makes that difference compared to the connectors that we see with

33:27 Play from here all of the Frontier Labs right now is that it's open source.

33:30 Play from here So they just keep getting added and added and added.

33:33 Play from here And as I was scrolling through yesterday, I'm like, wow, there's a connector for 1Password,

33:38 Play from here which I've been using for many, many years.

33:42 Play from here And I'm thinking that's a bold move.

33:45 Play from here Just give this thing access to your password repo and just being like, ah, here's my crypto.

33:50 Play from here Here's my bank account.

33:52 Play from here I have like so many ideas with workflows of just hooking this thing up to our CRM, to

33:57 Play from here our accounting system.

33:59 Play from here And that makes me a little squeamish.

34:01 Play from here Like I can't fathom just handing over my entire password vault.

34:07 Play from here All of this stuff should really be approached with caution first.

34:11 Play from here That's what I'll say.

34:13 Play from here Caution first with all of it.

34:14 Play from here But it's 100% worth playing with and just experimenting and trying stuff carefully,

34:20 Play from here carefully.

34:21 Play from here Carefully, carefully.

34:22 Play from here Yeah.

34:23 Play from here And again, you will end up on YouTube because, of course, you'll end up on YouTube.

34:27 Play from here And if you do, just be really careful with Claude, Anthropic in particular.

34:34 Play from here If you have a really valuable Claude account, like, I mean, I do.

34:38 Play from here I've been using Claude daily for three years.

34:40 Play from here A lot of history there.

34:41 Play from here There's a lot of history there.

34:43 Play from here I don't want to lose that account.

34:44 Play from here People are getting their accounts banned.

34:46 Play from here And you're talking about like these max plans where basically for $200 a month, it is quote

34:51 Play from here unquote unlimited, but you are violating the terms of service of hooking this up to

34:55 Play from here you.

34:56 Play from here Yeah.

34:57 Play from here And they're noticing.

34:58 Play from here And it's probably the most egregious offenders because the Claude max plan is supposed to

35:02 Play from here be basically unlimited.

35:03 Play from here So people were cranking, you know, 60 million tokens.

35:07 Play from here You're burning up entire like Midwestern lakes in data centers, like doing this stuff.

35:13 Play from here So that's now where I'm going to imagine all the water usage is coming from.

35:19 Play from here Your mileage may vary, but if I were to do this today, I would spin up a new Claude account.

35:25 Play from here I would actually probably use OpenAI because I think it's going to be a little bit more

35:28 Play from here liberal about this, but we need to get to why.

35:30 Play from here Okay.

35:31 Play from here We've just gotten really technical about all of this, but I guess for me, we talked just

35:36 Play from here a few weeks ago about this enterprise like knowledge graph idea and that we're already

35:42 Play from here getting beyond the point where the AIs can look at like SOPs and things like that in

35:47 Play from here your business.

35:48 Play from here And now in part, it's about filling in the gaps of all the other stuff that's going on,

35:55 Play from here which is why I wanted it to crawl over all of my transcripts.

35:58 Play from here So, you know, how many transcripts does my organization produce in a week?

36:02 Play from here It's a lot.

36:03 Play from here And being able to capture all of those, not just in individual workflows, which is what

36:08 Play from here we were already doing, call finishes, goes into a workflow, something cool happens, good

36:13 Play from here stuff happens out the other side, but also as this just foundational layer of information

36:19 Play from here in the organization that would be really hard for my brain to even accommodate how to sort

36:24 Play from here out.

36:25 Play from here And I want OpenCLAW to chew through that to identify an organizational identity, essentially,

36:33 Play from here and from a leadership perspective, make sure that I'm capturing all the things that I need

36:38 Play from here to be doing and also that other things in the organization are being captured.

36:42 Play from here So less time for me turning over rocks and more time for the bot to be turning over rocks

36:48 Play from here for me.

36:49 Play from here Right.

36:50 Play from here Right.

36:51 Play from here And that's really kind of getting to this whole chain sequence of things that it could

36:57 Play from here do where, and this is really a whole other discussion, but you know, we've talked about

37:01 Play from here workflow automation tools like N8n or Zapier, a bunch of other tools or Google's, Opal also

37:09 Play from here coming in.

37:10 Play from here I mean, there's a variety of things.

37:12 Play from here And one of the things that was making my brain spin a little bit this weekend was, does all

37:16 Play from here that kind of go away now as far as, you know, like, I mean, not only from like a level of

37:22 Play from here complexity because, you know, like, like you said with the, hey, now I can just, you know,

37:26 Play from here talk to my copier machine instead of pressing all these random buttons, instead of dealing

37:30 Play from here with JSON code or moving a bunch of nodes around a workspace, you just say, Hey, this

37:37 Play from here is what I want to do.

37:39 Play from here You've got these log creds, just go do it.

37:43 Play from here And it will go off and figure out how to do it and just make it happen.

37:46 Play from here So yeah.

37:48 Play from here Did you think about that too?

37:49 Play from here Like does all of that workflow automation just kind of go away at this point?

37:52 Play from here I mean, there are people who are listening to us right now have heard me do workshops

37:58 Play from here on it and, and how breaking these things down is sort of the next step that once you move

38:03 Play from here beyond chatbots and don't get me wrong, most organizations are still in like chatbot stage.

38:08 Play from here And then the obvious question is, okay, what do I do next?

38:11 Play from here How do I, now that my people are kind of chatting and they're using it for emails, that's cute,

38:17 Play from here but how do we turn this into something real?

38:19 Play from here And the answer is you start automating workflows and the best way to grapple with workflows

38:24 Play from here given the technology of, oh, I don't know, three or four months ago was to use these

38:29 Play from here automation tools like N8n.

38:31 Play from here And I still will be leaning on these for at least the next few internet minutes because

38:37 Play from here they are so much more predictable as far as getting the results that you want out of the

38:41 Play from here system. But yeah, if, if, if you as a manager understand what the right outputs are that

38:51 Play from here you want to get out of the system or what you want to get out of a person and can really

38:56 Play from here clarify that in a professional, if not a technical way, you increasingly are going to be able

39:04 Play from here to use tools like this.

39:05 Play from here I'm not sure this would be my go-to yet, but you could easily see it being an orchestrator

39:11 Play from here that like links in with Cloud Code or with OpenAI Codex.

39:16 Play from here I mean, this is going to happen.

39:17 Play from here So now it's like orchestrating all of the other technical layers, letting those technical

39:22 Play from here layers do their technical thing without necessarily have to drift into orchestration as

39:27 Play from here opposed to all of them trying to do all of the things.

39:30 Play from here Now you're, you're basically only limited by your imagination.

39:34 Play from here Yeah, yeah. So what you brought up, like your use case of taking all these transcripts and

39:41 Play from here I mean, technically you could have done something similar with an N8n workflow.

39:47 Play from here Like you could, you could make that work.

39:49 Play from here And I think you were probably just curious, like, can I scratch this itch a different

39:52 Play from here way and have this agent doing this for me?

39:55 Play from here And then also, you know, a lot of the things that have been coming out in the tech sphere

40:00 Play from here about all of this is the delightful surprise of watching this thing go off and solve

40:06 Play from here problems in ways that you didn't necessarily instruct it to solve problems.

40:11 Play from here Right. So that kind of novelty of it just came up with an approach I didn't even think of.

40:16 Play from here Whereas automation workflows are very rigid.

40:19 Play from here You're saying based on this input, I want you to do this.

40:22 Play from here Then I want you to call this model and figure this out.

40:24 Play from here And then I want you to take those findings and put them over here.

40:26 Play from here And it's a relatively rigid workflow with the intelligence layer being that part that's a bit more fluid.

40:32 Play from here But with these agents, really, the whole thing becomes kind of fluid.

40:37 Play from here So I don't know how much poking around you've done, Kevin, or people you've talked to.

40:42 Play from here I've heard a few things.

40:43 Play from here One developer shared that they let their mult bot just run over a whole weekend with access to their Slack and their GitHub, basically where they're stuffing all their code.

40:52 Play from here And by Monday, the bot didn't just summarize the messages in Slack and their dev channel, but it basically by itself identified this recurring bug that was mentioned that was a persistent issue in that dev channel, then researched a fix and created a new branch of code and submitted a pull request in the repo.

41:12 Play from here And, you know, the real shock there is that it performed the entire reasoning chain like it detected a problem.

41:18 Play from here It wrote the code.

41:19 Play from here It navigated GitHub and the user was sleeping and probably out at bars, you know, on Saturday and then came to this whole thing being fixed on Monday.

41:27 Play from here So that's one example of like what makes something like this so materially different than an automation workflow or an 8n.

41:37 Play from here So I'm going to put the brakes on here just for a minute because just yesterday I was a guest on, it was like a risk management podcast.

41:48 Play from here And, uh, you know, a lot of what we're doing right now.

41:50 Play from here Well, yeah, we're so half of my job is I'm like the bringer of AI fun.

41:54 Play from here And the other half of my job is like, well, but it's real, right?

42:02 Play from here Like, like what we're describing as far as these like fast and loose systems are not explainable.

42:08 Play from here And if you're dealing with, with high risk data and not to get wonky here, just, just think about anything that happens in an organization that has an irrevocable decision for humanity.

42:20 Play from here So if you're making a healthcare decision and insurance decision, hiring decision, there's a lot of stuff, loans, et cetera.

42:29 Play from here You have to be able to explain that because if you're held to account and you're just like, oh, then it went into a black box and good stuff came out the other side.

42:38 Play from here And with these, this, this, this, I've heard this bantered around about this idea of the, the fluid SAS.

42:45 Play from here And it's like every time a user runs something, it's basically inventing itself around them.

42:52 Play from here Like you're building a, you could almost picture like building the plane around the person as their, their, their prompt is happening.

43:01 Play from here But one day it has two tail fins and the next it has one.

43:05 Play from here And then, you know, it has a big wheel in the middle.

43:07 Play from here Like you can't, you can't run high risk businesses that way.

43:12 Play from here So when I think about things like N8N, I like to control as much of that as I can from what we call a deterministic perspective.

43:21 Play from here So it's going to do exactly what you expect it to every dang time, the same way it would have done three and a half years ago, if you were running it through make or Zapier or whatever.

43:30 Play from here It's going to pull data from somewhere.

43:31 Play from here It's going to run it through some code.

43:32 Play from here It's going to transform it.

43:33 Play from here It's going to move it around.

43:35 Play from here Great.

43:36 Play from here But focus on that one probabilistic step that's in there as far as where the AI is being applied and solve that problem really, really well.

43:46 Play from here And I think most organizations are going to appreciate that approach and being able to be very dynamic about how you're solving that one box is great.

43:54 Play from here But I don't think that you should be building the plane for all of the other pieces around.

43:59 Play from here it, unless you have a pretty good reason for doing that.

44:02 Play from here Thanks for bringing me back to reality, Kevin, because you're a hundred percent

44:05 Play from here right, and especially people inside are running big companies.

44:09 Play from here Yeah, this isn't really ready for what I'd call widespread commercial.

44:15 Play from here Not even, not even, not even close.

44:18 Play from here But I found myself in this headspace when I really started diving into some

44:23 Play from here of the workflow automation and then as a tool of, wow, I'm seeing stuff on

44:29 Play from here X or Instagram where, oh, that's interesting.

44:32 Play from here I never really thought of that being a problem I could solve with this.

44:35 Play from here And your imagination starts being the limiting factor for what's possible.

44:39 Play from here And so you put that on steroids and that's kind of what we're getting into now.

44:43 Play from here Like your imagination really does start to be the limiting factor.

44:46 Play from here And when you start hearing about things like, I mean, when you can just connect

44:50 Play from here to anything and this will just basically have the agency to go figure something

44:55 Play from here out, you know, an example is a voice agent, right?

44:58 Play from here Like you can, you can, you can go and create an 11labs account, which

45:02 Play from here will give yourself an AI voice.

45:05 Play from here And then you give your, your, your bot access to that 11labs account and it

45:12 Play from here can go and talk on behalf of you as well.

45:15 Play from here So, you know, one of the other little stories I heard was somebody letting

45:19 Play from here their bot like go handle a car price negotiation.

45:22 Play from here Did you hear this one?

45:23 Play from here It was, I heard one about a restaurant.

45:26 Play from here The restaurant woman's, yeah, yeah.

45:27 Play from here Like, oh yeah, yeah.

45:28 Play from here Here's my resi account.

45:29 Play from here You have, here's my log creds.

45:31 Play from here Just give me a reservation at this restaurant.

45:33 Play from here And then, you know, wake up in the morning.

45:34 Play from here I got the reservation, but that's weird.

45:36 Play from here Like looking at the history of tasks, there were no reservations.

45:39 Play from here So what happened?

45:41 Play from here Oh, well it went, logged into 11, called the restaurant and then made up some

45:44 Play from here whole story about why this person had to have this table at this time and basically

45:48 Play from here negotiated a table on their behalf, which is wild.

45:51 Play from here I don't even know if that one is true.

45:53 Play from here That's one of the problems.

45:55 Play from here It's horrible, right?

45:56 Play from here That, I mean, it's, it's, but it's like, yeah, like, so technically it's

46:01 Play from here like, yeah, that, that could totally work.

46:03 Play from here I mean, a lot of people are having a hard time telling the difference now between

46:05 Play from here a human and these different voice agents.

46:08 Play from here So yeah, like this guy just basically said, yeah, send these emails and we'll

46:13 Play from here figure out what's going to happen.

46:14 Play from here But the bot took it upon itself to then, you know, research on Reddit.

46:19 Play from here What are people actually paying for this specific vehicle?

46:21 Play from here And then it found a lower price than a dealership three towns over.

46:24 Play from here And use that as leverage using his voice agent to get a $4,000 discount.

46:31 Play from here Like it played a hard ball better than the owner could have played hard ball.

46:35 Play from here You know, which is, which is wild.

46:37 Play from here Wait until, wait until the bots are negotiating with the bots too, right?

46:41 Play from here I mean, we're not too far away from that, right?

46:42 Play from here Like that's really a thing.

46:44 Play from here That's, I mean, look at Multbook, right?

46:47 Play from here It's bots talking to bots at scale, you know, learning from each other.

46:51 Play from here Communicating different ideas.

46:53 Play from here That's it's that's already happening at scale.

46:55 Play from here So what you're talking about is really kind of around the corner, right?

46:58 Play from here Or, okay.

46:59 Play from here So like in just personal social stuff.

47:02 Play from here Um, so I was reading about one user who gave the bot access to their WhatsApp

47:07 Play from here and said, be helpful to my wife, right?

47:10 Play from here The bot began just proactively texting her good morning and good night.

47:15 Play from here And it even answered a question about their kid's school schedule

47:18 Play from here while the husband was at the gym.

47:20 Play from here And it was so convincing that she had no idea, but then it did start to like

47:24 Play from here make up some stuff about school events.

47:26 Play from here And she's like, wait a minute, that's not even true.

47:28 Play from here Right.

47:28 Play from here So, I mean, again, like this stuff is really interesting and fun to play with.

47:32 Play from here And it is giving us these little glimpses as to, okay, well,

47:37 Play from here what's, what's around the corner.

47:38 Play from here How is this going to change the way we work?

47:41 Play from here Uh, how far along until, you know, there are companies that are just stacks

47:46 Play from here of Mac minis on a desk and somebody orchestrating agents, right?

47:50 Play from here Like that's something that sounded really sci-fi.

47:53 Play from here I don't know, a month ago.

47:54 Play from here It's not a real stretch in my imagination to, to be there in a

47:59 Play from here relatively short amount of time.

48:00 Play from here It's wild.

48:01 Play from here Okay.

48:01 Play from here So again, orienting around the practical, you out there, you need

48:07 Play from here to know what you're trying to fix.

48:08 Play from here And there's so much stuff you can play with.

48:11 Play from here There's so many things that you can do, but most people are really scarce on time.

48:15 Play from here So how do you identify the things that are assailable and doable?

48:21 Play from here And really start with like friction points in your organization of things

48:25 Play from here that people don't like to do, or things that slow things down, or a question

48:29 Play from here that we often ask is what's the most expensive bad human decision that

48:33 Play from here you make in this organization.

48:35 Play from here And often it has something to do with quoting.

48:37 Play from here A bad quote leads to, you know, bad gross margin at the end of the day.

48:41 Play from here So how do you fix those things?

48:43 Play from here So make sure you're focusing on the right problems and start listing them

48:47 Play from here down and sort of categorizing them.

48:50 Play from here And then realize that there's no just defined path that I wish there were, but

48:55 Play from here you can solve problems in a whole bunch of different ways.

48:58 Play from here You can do it through a workflow builder like N8N and combine that

49:03 Play from here with Claude to help build the flow.

49:05 Play from here Or you can try and vibe code a solution.

49:08 Play from here We're not going to get there today, but vibe coding has gotten so much

49:11 Play from here better in the last few months.

49:13 Play from here Last few days.

49:16 Play from here Like we're now effectively using lovable all day long to produce real solutions,

49:22 Play from here both internally and externally.

49:23 Play from here Whereas before it was sort of, sort of solutions.

49:27 Play from here Now it's like legit stuff that you can build.

49:29 Play from here Or maybe it's an orchestration agent like this, or maybe

49:32 Play from here it's diving into Claude code.

49:34 Play from here And again, that's all dizzying.

49:36 Play from here But find the problem first and then decide that it's something

49:40 Play from here that you want to solve.

49:42 Play from here That's I think the best advice I can give to people is find a problem

49:46 Play from here that you want to solve and solve it.

49:48 Play from here Solve one problem.

49:49 Play from here Don't try and solve two dozen problems.

49:51 Play from here And that's going to define your approach to a lot of other problems.

49:54 Play from here And if it means you're using N8N next year when nobody else is, who cares?

49:59 Play from here If it's solved your problem and it continues to solve your problem, let

50:03 Play from here it continue to solve your problem.

50:04 Play from here Find a new way to solve your problem for a lot less money too.

50:07 Play from here I mean, let's be real right now.

50:08 Play from here It's, we're in a weird time of kind of balancing the economics with some of the

50:13 Play from here quote unquote problems people are solving.

50:15 Play from here A lot of this is just, let me see how far I can push this and it's going to be

50:19 Play from here expensive and there isn't necessarily immediate ROI, like there's just a lot

50:23 Play from here of kind of exploratory play going on here.

50:26 Play from here But here's, here's a question for you, Kevin.

50:28 Play from here I've been thinking about this a lot lately and I think this will be helpful

50:32 Play from here for listeners to think of as well.

50:34 Play from here There's always the kind of question of when do you jump into some of these

50:38 Play from here different technologies, right?

50:40 Play from here The, when do you jump in and where do you jump in?

50:43 Play from here And I think this is a great example of in the span of one day, you know, you're

50:49 Play from here messaging me, oh man, I'm sitting here trying to figure this out and figure this

50:53 Play from here out and then I'm, you know, I'm poking at you while your, your kids are a lot

50:57 Play from here older than my little ones.

50:58 Play from here I'm like, I'm outside enjoying, enjoying the real world in the sunshine while

51:02 Play from here you stare into your computer.

51:04 Play from here Meanwhile, I did the exact same thing the next day.

51:06 Play from here But in that one day, so much had already changed that I probably cut my

51:10 Play from here deploy time down by 75%.

51:13 Play from here You know, just because of tools that were released, things that people

51:16 Play from here contributed and it was easier, right?

51:18 Play from here So it always brings up this question of when, when do you jump into the stream,

51:26 Play from here right?

51:26 Play from here Like what is, what is the right, like finding that balance of we've all got

51:32 Play from here finite time, we all have problems we want to solve.

51:35 Play from here So not only picking those problems, but finding the right time to jump into the

51:38 Play from here right tools to solve those problems with something that's moving so quickly.

51:42 Play from here It's, it's, it's a subject I really do think a lot about.

51:45 Play from here And I imagine, you know, you're probably talking to many organizations about as

51:48 Play from here well.

51:49 Play from here All right.

51:49 Play from here So in the back of my mind was I'm going to jump back in after a week, after Eli

51:54 Play from here spends a bunch of money and figures this out.

51:58 Play from here Buy seven Mac minis and spends the next big portion of his life

52:02 Play from here configuring a bunch of bots.

52:03 Play from here Well, I mean, yeah, obviously we're early adopters and obviously we're

52:08 Play from here enthusiasts and I felt like I had to do this.

52:11 Play from here Like I just had to force myself to do it.

52:13 Play from here And it was a really interesting experience to do it, but you're right.

52:17 Play from here That's, that's not something that, that most of our listeners should be doing.

52:20 Play from here They need to, they need to hang out and wait a little bit.

52:23 Play from here Yes.

52:23 Play from here It's so cool that we can solve all the problems.

52:26 Play from here And yes, people like us are telling you that, that, that, Hey, your competitors

52:29 Play from here are doing this and whatnot.

52:31 Play from here Your competitors are not light years ahead of you in most respects.

52:36 Play from here So waiting for a little bit of maturation probably makes sense.

52:41 Play from here Uh, I don't know.

52:42 Play from here And the other thing is part of this is the collective building culture that I

52:47 Play from here think has transcended the technical.

52:49 Play from here So for years there was stack overflow out there and poor stack overflow, which is

52:54 Play from here where developers used to get their answers is dead because the models sucked in all

52:59 Play from here the information from it.

53:01 Play from here And now why would you go to stack overflow knowing full well, that you can

53:04 Play from here just have your record, basic, basic GPT question and get the answer.

53:10 Play from here And it's been liberating for those of us who I can't code, right?

53:13 Play from here That's not my, that's not my background and it was never my

53:15 Play from here interest, but I'm pretty technical.

53:17 Play from here So now I've been liberated to solve problems at sort of a macro level.

53:22 Play from here And what's amazing about right now is like, I mean, Toby Lutke, the CEO of

53:30 Play from here Shopify, I mean, certainly know who he is, but he's in the same threads as

53:35 Play from here everybody else, like the dude spent his Saturday solving the same problems

53:39 Play from here that I was trying to solve.

53:41 Play from here And he is, he, you know, he's came up with a new file, like a 20 hour bender

53:47 Play from here of like trying to figure this stuff out.

53:49 Play from here It's completely wild.

53:50 Play from here This guy's a billionaire.

53:51 Play from here And, you know, he spent his Saturday, like in the same trench that I was in

53:56 Play from here trying to figure this stuff out.

53:57 Play from here And that's going to accelerate progress too, because it, you know, all of

54:02 Play from here this information is out there on X.

54:03 Play from here All of this information is out there on other forums and you know, who

54:07 Play from here has access to all of that?

54:08 Play from here The LLMs do.

54:09 Play from here So here we talking about the acceleration again of all of it.

54:13 Play from here I don't know, you were, you were trying to land on a reasonable, and

54:15 Play from here then I got really excited again.

54:17 Play from here But the, like, what is the answer?

54:19 Play from here What is the time period?

54:20 Play from here Because I don't want to tell people to wait for like, okay.

54:22 Play from here If you're a Workday customer, Workday is doing amazing things with automations.

54:26 Play from here You are probably, if you're that big a company to be, to have Workday as a

54:31 Play from here vendor, you're probably going to use their solutions, but most mere mortals

54:36 Play from here don't sit in that place and most mere mortals do not use vendors that are

54:40 Play from here really on the cutting edge of this.

54:42 Play from here So there is an inflection point here between waiting and having a

54:47 Play from here pain point that needs to be solved.

54:49 Play from here Yeah.

54:49 Play from here It's interesting because, you know, you mentioned, oh, my competitors

54:52 Play from here are running away from me and this is crazy.

54:55 Play from here And they're, listen, this is, this is in one way been like a real

55:01 Play from here re-leveling of the playing field, right?

55:03 Play from here Where you are seeing billionaires in the same forums that you are

55:06 Play from here trying to find information and asking the models the same question.

55:09 Play from here So on one hand, it's never been a better time to be a tinkerer, to

55:13 Play from here just learn stuff and play with stuff.

55:15 Play from here And again, if you are far along enough in your AI journey to be routinely

55:22 Play from here getting the things that you're looking for out of these different models,

55:26 Play from here just by chatting with them, you don't honestly have to be super technical to

55:30 Play from here get a hosting your account, to spin one of these up, to play with it, just by

55:33 Play from here having a conversation with GPT or Gemini, you don't have to be super technical.

55:37 Play from here Like you can get there.

55:38 Play from here Now, should you go and start deploying agents in your business tomorrow?

55:41 Play from here Probably not, or even spending your time trying to figure out how

55:46 Play from here that would work or how it'd be useful.

55:47 Play from here But there's a difference between that and just not sleeping on this and playing

55:52 Play from here and exploring and just keeping your eye on a very quickly moving ball.

55:57 Play from here That's, that's what I would say.

55:59 Play from here And we'll be watching how you perform with said ball of your eyes.

56:02 Play from here Yeah, yeah.

56:04 Play from here I, I know here, here, here we are again, looking at each other in our little box

56:09 Play from here of lights live 20 minutes away from each other.

56:12 Play from here I don't think I've seen you in months now, but I've got a lot of thoughts on this.

56:16 Play from here A lot of ideas.

56:17 Play from here Um, it's, it's, it's, it's a, it's a very, very interesting

56:21 Play from here time to be alive in tech.

56:22 Play from here It really is.

56:24 Play from here All right.

56:24 Play from here I think that's it.

56:25 Play from here I think that's it.

56:26 Play from here All right, man.

56:27 Play from here Good to see you.

56:27 Play from here That's the pod.

56:28 Play from here We'll talk next time.

56:34 Play from here Thanks for listening.

56:35 Play from here If you'd like to send us a question to answer on the show, email me at

56:39 Play from here at spark six.com.

56:40 Play from here That is S P a R K.

56:42 Play from here The number six.com or Kevin at Kevin at ascend labs.ai to learn more about AI

56:49 Play from here strategy and training for your business.

56:50 Play from here Visit Kevin's company, ascend AI labs at ascend labs.ai.

56:56 Play from here And to learn more about design and development of custom AI solutions,

57:00 Play from here visit my agency at spark six.com.

57:04 Play from here You can also sign up for my free newsletter, AI framework Friday by

57:07 Play from here visiting spark six.com forward slash newsletter.

How to cite this episode

Credit the named speaker and AI at Work by AscendAI. Link to the episode or the passage you quoted.

“Ep #12 1.6 Million AI Agents Built a Social Network, And Humans Could Only Watch.” AI at Work by AscendAI, 2026-02-03. https://ascendlabs.ai/podcast/1-6-million-ai-agents-built-a-social-network-and-humans-could-only-watch/

Recent episodes

View all episodes
×
AI Solutions Graphic
Free Resource
Get Started Now