All right. So, I want to talk about what

happens next. And this is with respect

to Open Claw and Molt Book and Rent a

Human and all that that kind of fun

stuff. So, if you're not if you're not

up to speed, here's a 202 catch you up

to speed. Uh, a few weeks ago, someone

or actually guess it was end of 2025.

Someone created OpenClaw or what was

called Claudebot at the time. Fully

autonomous uh agent that, you know,

works for you around the clock. uh runs

on scripts and cron jobs and it wakes up

at a you know regular cadence to go do

stuff autonomously for you and then

someone made claude book um or no claude

book whatever it was called um basically

read it for uh sorry mold book there we

go my brain was like no that's not right

mold book so basically read it for these

agents now that immediately became a

cess pit for cryptog and humans writing

posts and that sort of thing. But it

made open claw very popular. And then

someone extended that with rent a human,

which is basically uh AI agents can go

pay a human to do something for them.

Now what's happening? So what is what is

legitimately actually physically

happening in the real world is yes mold

book was full of grift. Um but it shows

a path forward. At the same time what is

actually physically true is that lots

and lots of people hundreds of thousands

of people if not millions of people are

using openclaw around the world uh to do

real work. And you know, people talk

about it and and the conversation is as

sharply divided about this kind of agent

as it was for chat bots when chat bots

first got big cuz you still see see

people out there saying, you know, I

don't I don't understand a legitimate

use case for chat GPT and it's like,

okay, buddy, you know, you're you're

just a lite and you're going to live in

your cave. I remember when when chat GPT

first launched and I was still on the

OpenAI forums, there were guys going

around saying, "This is just Eliza. Tell

you need to prove to me that this is

like people get so huffy and indignant

and like angry when the paradigm

shifts." And we're seeing the same exact

thing. I think that this is why it took

me a while to like figure out how to

talk about this is because the reaction

is so sharp around openclaw and these

this idea of autonomous agents that it's

it's like that basically we've had a few

paradigms. So from my perspective having

been in this since been in this like

full-time since GPT2 um so we had we had

the original LLMs which were literally

just autocomplete engines. They weren't

even chatbots yet. So that was paradigm

one. Then paradigm two was the instruct

models where I was like okay well

instead of having to give give every

single time you give a a prompt example

let's just go ahead and pre-train the

models to follow simple instructions.

Well once we had that it was a hop skip

and a step to get chat bots cuz then the

instruction is just be a chatbot with

you know x y and z personality. So we

had plain vanilla, we had instruct

aligned, then we had chat bots. And chat

bots was the first like you could say

instruct aligned was like paradigm 1.5

because it was still basically just an

autocomplete engine, but it was an

autocomplete engine that was designed to

follow at least a single instruction

relatively well. The chatbot was a was a

fundamentally different UX. So chatbot

was paradigm 2. And I didn't even use

chat GPT for the first few months cuz

I'm like I don't care. Like that's just

a chatbot. such as a distraction for

this this model. But of course, time has

proven that once you add on tool use and

you add on reasoning and a bunch of

other things, then the chatbot actually

becomes semiautonomous.

You know, like I still regularly use

chat GPT Pro when I'm doing heavy duty

math research to model like, you know,

how many how many jobs were eliminated

in 2025. And by the way, cross-checking

across multiple AIs, it looks like

artificial intelligence ultimately

either destroyed or avoided about

200,000 to 300,000 jobs in America alone

last year. Um, so anyways, just wanted

to throw that little factoid out. The

official number is 54,167

or something like that. Um, so the the

real number and the way that you detect

this is there's kind of two ways you

look at number one, you look at um

excess deaths. So during co times and

I'm sorry for this tangent but this is

actually kind of important and it's not

I don't I don't want to do a whole video

on it. So anyways during co times we did

excess deaths which is basically all

else being equal you look at how many

extra people died in a given time and

place and you say okay well the primary

difference is COVID. So then you

attribute those excess deaths to CO

exposure. So we did the same thing for

jobs which is all else being equal

looking at things like inflation data

and um and in and interest rates and

those sorts of things. You can generally

anticipate how many layoffs should have

happened and then so then you say okay

well how many excess layoffs did we

have. Then the other thing that you can

do is you can look at it the opposite

direction which is labor growth. So then

you compare things like GDP growth to

the actual number of jobs created and

you look at the at the difference

between those. And when you correct for

everything else like you know Doge

layoffs and whatever else was going on

last year um then you get somewhere

between like a 100,000 and 350,000

uh jobs were either destroyed or

avoided. And this this actually goes

back to an AI layoff is not necessarily

just someone getting fired handed a pink

slip saying AI was responsible for this.

It is that new jobs are not being

created. Uh, so anyways, um, that's an

example of what I use ChatGBT Pro for

and Gemini Pro and um, I don't use

Perplexity Max anymore, but I have a

Claude Max. So, I use all of these

different tools, but they're just

waiting for me. So, the the next big

paradigm shift has been agents. And

people have been trying to crack the nut

of agents for years now. like the ACE

framework if you remember like ACE

framework the Raven framework um we've

been trying to crack this nut for a

while and then dude with open claw

figures it out so here we are now what I

will say is that adding reasoning and

tool use was kind of paradigm 2.5 cuz

that was still fundamentally the same

form factor as chat bots so it was chat

bots with a little bit of extra bolted

on but then those chat bots with a

little bit extra bolted on has evolved

into openclaw So, Paradigm 1 was just

autocomplete engine and then we made

them a little bit better until they

evolved into chatbots. It's like

Pokemon, honestly. So, you know, the the

the the plain vanilla original Pokemon,

you know, that you get right at the

beginning of the game is just the

autocomplete engine. Stage two is the

chatbot. Stage three is the agent. Now,

the agent allows for a lot of emergence.

So you've probably heard uh everyone

from me to systems thinkers like Daniel

Schmokenberger and everyone else even

the AI safety people we talk about

emergence. So emergence has a few

connotations in in pure AI the idea is

that new abilities emerge as you get

certain levels of sophistication data

parameter counts architectures those

sorts of things. So like theory of mind

seems to have emerged. Um, but what has

what has also happened is that once you

know to look for an ability in a

language model, it's usually been there

all along. It was just not prominent

enough to be useful. So like theory of

mind, GPT2 has planning and theory of

mind. It's not very good, but it's in

there. So then it got better with GPT3

and people still didn't even know to

look for it even though a lot of us were

talking about it. Um, and then once you

get to GPT4 and GPT5, then people are

like, "Oh yeah, it's clearly got good

enough theory of mind." um and planning

and reasoning. It's always been there.

It just wasn't that good. So now it's

there. It's better. It's useful. And in

fact, in in the case of theory of mind,

um modern like frontier models are

generally better than the average human

at theory of mind. Um so that's that's

one example of emergence. But another

version of emergence because that's

within a single system. um within in in

larger systems in complex systems or or

um well yeah I guess in in systems

theory emergence can be like emergent

gameplay is an example of emergence

where you have a game that is

sophisticated enough with enough rules

that people can kind of make their own

games in this. So um some examples are

like Minecraft and Roblox and those

sorts of things where there's enough

different game mechanics. you have

enough you have a you have a

sophisticated enough alphabet of game

mechanics that you can get spontaneous

uh new forms of gameplay. So all the big

games like Fortnite and stuff where

there's a building mechanic, a survival

mechanic, a a fighting mechanic and

those sorts of things, you get emergent

gameplay. Um so that that is that is a

real life example of emergence. Now when

you apply that to things like agents, so

now you have agents interacting with

each other and they're interacting with

each other, you know, locally, they're

interacting with each other publicly,

they're interacting with humans, they're

interacting with businesses. That's

where you get another layer of emergence

because the thing is is with chat bots,

you had a very constrained environment

where a chatbot was interacting with

you. You were pretty much the the

primary variable. And then of course you

know it it had tool use and retrieval

augmented generation and reasoning

abilities as kind of the three primary

food groups. Obviously that's they have

lists of tools. Um but they can get

information from elsewhere. They can run

programming and other tools and then

they have reasoning abilities. So those

are the three big variables. But it was

still a constrained environment because

it was just going to do a process. You

know they could think I think my longest

chat GPT pro was like 58 minutes or

maybe just over an hour. and then it's

going to spit out the results and then

wait for you. So that there's a there's

a very definitive time step and it's

based on the human. However, whenever

you have complex systems, you introduce

more and more variables which which

ultimately creates more chaos and I mean

this in like the strict like the the the

legitimate like chaos theory because

with a single chatbot doing a single

thing, what we're trying to do with

safety and alignment is reduce the

chaos. Meaning we know that every time

you use Claude Max or G Chat GPT Pro or

Gemini Pro or Gro Heavy, the input

processing and output cycle is going to

be very predictable. And what I mean by

predictable is it's not going to go off

the rails. It's not going to do any

financial harm. It's not going to

accidentally teach you how to make drugs

or nukes or that sort of thing. And so

that's the time step. So we've basically

been operating up until now on an

individual loop. So the loop is human

puts something in, the AI goes and does

something and then gives the human

output. So that's the input processing

output loop and that's the loop and it

just keeps looping. But what openclaw

does is it makes the loop not dependent

on humans. So then not only do you have

incremental time steps that are not

limited to humans, you have incremental

time steps that are influenced by other

agents and other environments. And that

introduces basically irreducible

complexity or or chaos. And so this goes

back to the original work that that we

did and I say we because we had a very

large community working on Gateau. So

that's the global alignment tax taxonomy

omnibus framework which we we realized

you know anyone working on cognitive

architecture as of three plus years ago

realized and I remember one of the one

of the guys that we collaborated with he

said I realize that these things are

going to be talking to each other more

than us very soon and that has turned

out to be true. Now, the reason I'm

bringing all of this up is because from

a risk and compliance and safety and

liability standpoint,

um it like openclaw is intolerable. And

what I mean by intolerable is that no

frontier lab is going to touch this

anytime soon. Um no Fortune 500

enterprise is going to touch it anytime

soon either. one of the biggest barriers

of adoption for even professional-grade

agentic frameworks and I I have Silicon

Valley founder friends um who who are

who are building these kinds of things

is number one interoperability. So like

what a lot of people expect an agent is

going to be is just a digital human that

you drop in and there are some people

working towards that where you just put

it in a in a a virtual machine give it a

virtual Windows desktop and away it

goes. But um the the the native space

that these things live in is more like

OpenClaw where it's all terminals, it's

all command lines, it's all API calls.

So rather than saying, okay, well, you

need to um have a keyboard, a mouse, and

and a screen and a virtual screen that

you have to look at and interpret, it's

like, why don't I just give you the the

information directly? Um so they're

they're they're not guey native,

graphical user interface native. It's

actually kind of like back to old

school. And this is why whenever you

watch people that are like hacking on

OpenClaw and stuff, they have like four

monitors up and it's all terminals. It

looks like the Matrix, right? Um, and

that's because their native environment

is textbased, meaning that the terminal

output from, you know, command line or

APIs is natively what they they read.

It's there's actually too much

information. There's too much noise if

you give them a screen, if you give them

a camera. Uh so

that is all right that was a little bit

of a tangent but I I felt like I needed

to explain it but the the overarching

point here is that from a security so

there's there's a few ways to look at

this. So if we if we put our you know um

Fortune 500 hat on we say all right so

you got cyber security and they say well

what are the risks and it's like oh well

you give it root access to a virtual

machine and it can have prompt

injections from from um infected skills

that it downloaded online and so cyber

security is going to say banan it say

this is not a product this is

functionally malware um and and in many

cases that is one of the best ways to

characterize openclaw from a cyber

security perspective. I'm not saying

that it is literal malware, but I'm

saying having worked in Fortune 500

companies, numerous of them. I'm telling

you that is how cyber security would

treat it. And honestly, as an as a

former uh automation uh infrastructure

guy, I would say like, yeah, you can't

trust this because you if it if it

touches one of your routers or one of

your switches or one of your servers or

one of your storage arrays and it runs

the wrong command, it shuts everything

down and you're losing, you know, in

some cases tens of millions of dollars

per hour on top of reputation damage and

on top of legal liability. So, it will

take a very long time. And when I say

very long time, the fastest I could see

many um Fortune 500 companies deploying

something like not OpenClaw directly,

but a successor to OpenClaw is maybe 18

months. And that is just because that's

how long it takes for you to do an

infrastructure audit, a cyber security

audit. Um they will have like toy

versions set up to play with. And you

know, cyber security team, they'll

they'll be monkeying around with it

already. But one of the other things

that you absolutely need to have is

executive buyin at the very top. And and

when I say at the very top, I don't mean

your chief technology officer. It has to

come from the CEO and ideally the board

of directors. Um the the the companies

that we are seeing that are and this is

this is all organizations. It doesn't

have to be a for-profit company. It

doesn't have to be a Fortune 500

company. It can be a government. It can

be a mid-size company. It can be an

enterprise. It can be a small company.

The only companies, the only

organizations that are successfully

making this this pivot is when the owner

or whoever is the highest stakeholder,

whoever it happens to be, whether it's

the CEO, the owner, board of directors,

the governor, whoever it is, if they

issue an edict saying we are going allin

on AI and they are the ones leading the

charge. And the reason is because AI is

so fast moving and it's so scary and

organizations are so riskaverse that if

you don't see the head honcho constantly

saying oh look what I did with chat GPT

then everyone else is going to be like I

might use chat GPT a little bit but I'm

going to hide the use and I'm not going

to do that kind of thing. my wife's

company, it's not a company she owns,

it's a company she she contracts for.

The owner and CEO and founder, all the

same person, on their weekly meeting

says, "Tell us what you used AI for."

And so, he's created a culture of, "Hey,

we're we're treating AI like a first

class asset." It's like, "Imagine

imagine you didn't have computers and

you didn't have internet." And it's

like, "Oh, well, the factory has never

needed computers or internet, so we're

just going to kind of ignore it." That's

how most companies and organizations are

reacting. Um, but the companies that are

saying, "Hey, we don't know how to use

this, but go play with it." Um, and you

know, it's like play with it until you

go blind. And if you know, you know.

Anyways, this is a little bit of a dirty

joke. Um, the idea is that you need

executive sponsorship. And when I say

executive sponsorship, I mean like

wholehog buyin. They need to be the ones

leading the charge. In fact, our

consultation business, our side business

that we're working on, we have decided

that we will not work with any clients

for whom like it's not coming from the

top because the thing is is in the

consultation space you might have um you

might have a company or a manager or you

know even a sea level we have we have

seen cases where the sea level says we

need to go allin on this but you know

the the chief financial officer chief

head of legal counsel, chief of HR,

they're all either ambivalent or

skeptical. And so the CEO is like,

"Well, you know, do what you can, but

it's not a top priority." If the CEO

says it's not a top priority, we walk.

Like, we we just we know that they're

not ready. And the reason I'm telling

you all of this, and I know that most of

you have never been in a Fortune 500

company. Most of you have never worked

for and have no interest in working for

a Fortune 500 company. The reason I'm

telling all of you this is because one

that's like they did that for 15 years.

But then also so that you are aware that

the gulf between what you know is

possible on the cutting edge and where

most people are is actually getting

wider because we still have people

saying like well I don't really see a

role for chat bots in enterprise. I

don't really see a role for chat bots in

government. and they're all so

riskaverse and they're so slowmoving,

it's going to take years for them to

really come around. Meanwhile, the ones

that are rapid adopters are going to be

pulling ahead. And honestly, when this

in in hindsight, imagine you had a

company that said, "Well, we don't

really get this internet stuff. We don't

really get this personal computer stuff.

We don't really get this cloud stuff."

What happens? A lot of them go out of

business. Look at Borders, right?

Borders is is one of the primary

examples where they said books that that

that's not changing. People like having

a physical book in their hand. And here

comes Jeff Bezos with his, you know,

warehouse robots and bulldozes, that

whole thing. Barnes & Noble managed to

survive somehow. I don't even know how

cuz when was the last time you even went

to a Barnes & Noble, but Borders is long

gone. And so what I'm what I'm telling

you is that the companies that survive,

that decision is being made now. But

it's like it's kind of like, you know,

you you see in a war movie or or

whatever where it's like, "Oh, he's

already dead. His body just hasn't

caught up to the fact yet." I don't

remember what movie that was from, but

it it's it's a trope, right? It's like

the there are already zombie companies.

There are dead men walking out there and

it comes from the top because of their

attitude towards artificial

intelligence. If they're dragging their

feet on chat bots, they're they're

hardly using any language models at all.

you know, they might still be using um

one of the one of the first use cases

that has diffused out is using small

language models to do things like help

with automatic routing of tickets and

that sort of thing. You know, the in in

business theory, there's a there's a

concept called forgivability, which

means what's the cost of doing it wrong

and how difficult is it to reverse? And

so, if you route a ticket to the wrong

person, that has basically zero cost.

It's like it's going to cost someone a

human 5 minutes to route it back to the

right place, but no lives are lost, no

money is lost, and there's no legal

exposure. So, you go through who's

everyone that you need to convince on

the board or, you know, at city council

or town hall or the governor's office

or, you know, the seauite. So, you've

got legal, you've got finance, you've

got HR, um you've got cyber security. So

even when the tech nerds over here say

guys this is very clearly the way of the

future you're going to get a lot of

resistance and in some cases it is

rightly so because to be fair a lot of

us tech guys want to move faster than

the rest of the organization is even

capable of moving fast or is or is would

even be um would even be uh wise to move

that safe. And the reason is because you

really have to dot your eyes and cross

your tees and justify, okay, if we roll

out these chat bots, which by the way,

Microsoft is charging $40 per month for

co-pilot. And you know, that doesn't

sound like much. And from my

perspective, the argument is very

simple. It's like in that for in that,

you know, for for every single seat,

will you get $40 more value to the

organization than that cost? So, like

the ROI for for people who use chat bots

is very obvious where it's like, hey,

you're paying someone effectively $50 to

$60 an hour. All you have to do is get

an extra hours worth of value to the

company per month out of that employee

and the cost is justified. But that's

not even how CFOs think. CFOs say, "Yes,

but one, how do we track that? And then

two, what if they just get the same

amount of work done faster and then

they're lazier?" So those are very real

considerations, but for most people like

especially smaller organizations where

it's like, yeah, we're all using chat

GPT or Gemini or whatever all day every

day and we are 10 times more effective

as a team. That is much easier to

achieve when you have a 20 or 30 person

organization rather than a 20 or 30,000

person organization. And so what what

we're and I'm reporting from the front

lines. This is what we're literally

seeing out there in the enterprise space

is that you have um you have like how am

I trying to say this? You you have a lot

of job roles where chat bots really

don't help that much. Um and then and

and that's a true thing. Like if you

spend most of your day in a truck, you

know, moving heavy materials around or

meeting with people face to face, you

might use a chatbot at the end of the

day to help write reports. Now, to be

fair, if you spend four or five hours a

week writing reports and the chat bots

can help you do it in 30 minutes, so

then you can spend more time face tof

face with clients, with customers, with

suppliers, then again, that should be a

no-brainer. But that's not how CFOs

think. And then of course your chief

legal counsel is just like you know

here's here's an example that I learned

recently is that uh insurance will not

underwrite AI right now. They don't know

how to price it in. Um and so well if if

insurance won't write it then there

that's an an an intolerable amount of

financial risk andor legal risk. Uh so

then legal says shut it down. Don't use

it. Now, of course, there's often a lot

of hypocrisy because finance might be

using the chatbot as shadow IT. Legal

might be using the chatbots as shadow

IT. HR might be using the chatbots as

shadow IT. So, then you have this other

problem. And this is where your your C

your CISO, your chief information

security officer comes in. So, the CISO

says, well, we have this problem with

with shadow IT. Legal's using it and we

can't tell them to stop. HR is using it

and we could probably tell them to stop.

um all of it is using it and we can we

can definitely you know lean on on the

CTO to have a lockdown policy but people

are going to be trying to use it anyways

regardless of what we try and say or do.

So then there's almost kind of a forcing

function or a binding function because

these things are just not ready and and

when I say these things I'm specifically

referring to like the agents because

what the journey that we're on is

there's the kind of the three phases.

There's the plain vanilla autocomplete.

There's the chat bots. And then there's

the agents. So all of this is to is to

show that it's going to take longer than

is remotely reasonably

uh uh I guess considerable or what is

what is reasonable to deploy these

things because often the limitation is

not technology

particularly with fastmoving new stuff.

The limitation is the rest of the

organization. And so like I have I have

a a friend a professor friend who also

consults for the state government and

it's like you have to have so many

endless meetings and all of us tech

people get so bored with it. We're like

guys you are like 2 or 3 years behind

the curve. It's time to move on. And

this is not to say that like you know

the singularity is canled or that sort

of thing but if you're watch if you are

one of the people watching this video

you are at the tip of the spear like you

are on the bleeding edge. you're on the

cutting edge and you know what's coming

and you know that it's going to take

like you know chat GPT's been out for

what three years now and and people are

still arguing like oh I don't really see

any b business value for that it's like

okay well that's that's a you problem

friend likewise there's people there's

there's high-tech cutting edge people

out there who are honestly saying I

don't see a use case for openclaw I

don't see a use case for maltbook I

don't see a use case for rent a human

this is all silly and that's just not

how general purpose technologies evolve.

What we're seeing right now is like

taking electricity and and you know, use

case one of electricity was like light

bulbs, which is basically you just short

it out, right? If if you if you have a

if you have a current and you short it

out, it makes light and heat. So it's

like a light bulb is like the equivalent

of the autocomplete era. And then

electric motors came around and it's

like, oh wow, if you do some coils and a

stator and some magnets, you can

actually convert that current into a lot

of torque. Cool. But then you have more

sophisticated uses of electricity like

communication being one of the most

sophisticated things. And then

computation, all of that is still

predicated. So like communication like

technology so so radio, telephone,

telegraph um and and switched networks

that's like layer three that's like uh

iteration number three of of the uses of

electricity as a general purpose

technology and then computation is is

layer four. So like that's a fourth

order consequence of electricity. It was

not immediately obvious when the first

people said, "Oh, like if we put these

chemicals together with different

metals, then we get like a little spark.

Hey, cool. What does that mean?" I don't

know. Or if you know the first time

someone spun up a dynamo and it's like,

"Oh, cool." Like it can give a little

zap or a little arc. What do we use this

for? We can use it to make rocks think

like that was not obvious. And my point

there is that it's not obvious to go

from an autocomplete engine which is

just a token producer to then reshape

that token producer to a chatbot and

then give it reasoning and give it tool

use and then reshape that again into

autonomous agents. It was obvious to us

technologists. That's why I went that I

literally decided to go all in on AI

when GPT3 came out and that's how it

became my career because I said this is

very obviously the next big thing. But

the diffusion part this is what we're

in. We're in the long slog of diffusion,

meaning it's going to take time. It's

going to take experimentation. There's

going to be emergent risks. There's

going to be emergent benefits. There's

going to be emergent form factors. All

kinds of stuff. And this is why, you

know, like, okay, let's just use an

example for something that didn't pop

off, which is VR. So, the metaverse, VR,

XR, AR. The idea was that we have a new

primitive which is, you know, basically

a head-mounted device, so an HMD as an

interface to cyerspace. And so, you

know, if you go all the way back to the

1980s and and maybe even before with

particularly uh Japanese culture, manga

and and anime, the idea that you'd have

a headset or some kind of headmounted

device and that you'd dive into the

cyberspace and the etherwebs and

whatever else, that [snorts] was just

people thought it was a foregone

conclusion that that was the future. We

actually have those headmounted devices

now and nobody uses them. Um so there

it's like okay we we invented one new

technological primitive and it didn't go

anywhere. So it's understandable that

people say okay well you have you have

you know tok you have a uh you next

token prediction and that's a new like

technological primitive. So yeah some

people assume it's not going to go

anywhere but they're wrong just plain

and simple. And even though they're

wrong it's still going to take time for

this technology to diffuse out for it to

mature and those sorts of things. So, I

don't really mean to like rain on the

parade, but like when you when you

straddle two worlds and the two worlds

that I'm talking about right now is like

the frontier, like what's happening in

real time, what is actually physically

possible versus what the companies and

governments are actually going to do.

It's very depressing. So, [laughter]

that's all there is to it. Talk to

you'all later.