We certainly have gone beyond whether if

we will use AI, but rather how we will

use AI. But we still see AI adoption

curve lag behind expectations,

especially when it comes to looking at

them from the enterprise level. And

here's one way to dissect the industry

into three segments. AI applications,

agentic applications, and AI native. So

given this framework, I'll be covering

how AI adoption plays at each of these

segments and drill down on specific on

barriers in AI adoption. A quick shout

out to Woven sponsoring this video. More

on them later. AI applications solve

immediate problems that are meant to

help us with tasks. Applications like

Chacha PT, Perplexity, and Copilot fit

into this category. And you can use them

at your disposal without changing how

you get work done. You can use this tool

optionally to get something done or just

do it the old way without it. Totally up

to you. And typically these applications

have lower barriers to entry from you

since it's relatively non-invasive in

nature and require little upkeep and

little information from you or your

system. So AI adoption on the AI

application level is a lot higher since

the barrier to entry is extremely low

and they more or less help you get

things done faster on a task level.

Agentic applications are a bit more

invasive than AI applications because

they are autonomous by nature. The

barrier to entry is slightly higher

since they require more from you and

your systems to get going. And they're

meant to go beyond a simple task

completion tool, but deeper into the

workflow. And contrary to popular

belief, agentic applications are

actually pretty rare, mostly because

people either aren't ready for it or

they just don't want them because it

generally doesn't solve the right

problem. And finally, AI native, which

is a relatively new and uncharted

territory, but worth mentioning because

it's sort of like the golden snitch from

Harry Potter, where catching the golden

snitch is difficult, but if someone ends

up catching it, the game ends in your

favor, even if you are losing the game

in the first place. The idea behind AI

native is that everything starts and end

with AI systems front to back. So,

tasks, workflows, and systems, all of

it. And while you might have a lot of

skepticism towards AI native, since most

organizations try to adopt AI from

outside in, meaning they try to automate

existing tasks using AI applications,

then transition into agentic

applications to automate workflows and

then finally apply AI systematically.

But if you think about it from startups

that have the luxury of starting from

scratch, this fits into the narrative of

golden snitch analogy where if executed

right, it could be just as viable as the

other way around. Now keep in mind that

the topic is not about what strategy

works best in adopting AI but rather

looking at the current landscape of AI

based applications and how they are

being adopted by the public and

organizations to draw insights on what

AI adoption actually looks like. Let's

start with an application that we are

all familiar with which is Chacht. The

reason why Chacht is a really good

example to start with is because the

barrier to using Chacha PT is probably

lower than most AI applications that we

have today. And that's because the

learning curve is so low to start using

it. And you don't need to connect chach

with other systems to start seeing

values from it. You just log in and

start asking questions. But just as easy

as it is to start using chach, a lower

barrier to entry typically means lower

return on value. Meaning if you look at

it from return on investment for chacha,

it's somewhat fixed to the amount of

work going in to get the value out. For

example, if you wanted to use Chachip

Piti to write an essay, you put in all

the work in defining the scope and

parameters of the essay to get around

70% of the work done by Chachi PT. So

the value that Chachip brings is

somewhat tied to the effort that you put

in since it's an isolated system that

only helps you with the tasks that you

give, which in this case is writing an

essay. Another example is coding agents

like claude code or client. Even though

these applications are advertised as

agentic applications, in reality we use

them to solve specific tasks like

writing code, modifying code, and code

reviews. So here, even though cloud code

carries out tasks agentically where it

autonomously reads and writes code in

your codebase, they are not a true agent

in a sense because they don't go beyond

a simple task completion tool. In other

words, completing tasks agentically is

not the same as agent autonomously

completing tasks. And meanwhile, we have

news articles like the Anthropic CEO

saying that AI could be writing 90% of

their code within six months. Meta also

saying 20 to 30% could be written by AI

and Microsoft also saying that 30% of

the code in Microsoft could be written

in AI. So looking at these figures, it

certainly feels like AI is about to

disrupt the entire coding industry. But

the reality is that these projections

are still focusing on a task level.

Meaning because AI applications are

getting more and more effective in task

completion like writing new code,

modifying existing code and reviewing

code, the AI applications are more like

task delegation tool where we ask AI to

complete a specific task and delegating

tasks are important especially when it

comes to hiring people. Which is why I

wanted to do a quick description of

Woven sponsoring this video. I've been

looking to hire a software developer in

my previous company and one thing I

always found was that candidates always

had different skill sets and some people

were really good at code reviews and

others were good at system debugging and

now with AI agentic programming. So

coming up with coding evaluations for

each role took a lot of time and effort

to build scenarios and give feedback. It

just wasn't fun for everyone involved in

the process. Woven is a humanpowered

technical assessment tool that makes

hiring streamlined. So, if you're

looking to hire engineers, Woven is

offering 14 days free trial with 20% off

of your first hire. Check the link in

the description. So, while we truly are

seeing complete disruption on task level

where tasks like writing code, reviewing

literature, modifying existing code, and

code reviews, things start to look

differently when it comes to agentic

applications. Even though AI

applications can certainly complete

tasks agentically, agentic application

is a different paradigm where it goes

beyond task completion but rather

workflow automation and autonomous

execution. Let's think of this scenario.

Let's say I wanted to sell you an AI

calendar app, but I gave you two

different explanations. First, this is

an AI application that helps you manage

your calendar. And second, this is an AI

application that manages your calendar.

Although the wording is similar, the

impact is quite different. The first

statement, the AI application that helps

me manage my calendar. I would be more

inclined to use it since I don't have to

change how I already manage my calendar,

but it could help with calendar related

tasks like adding appointments and

changing schedules. In other words,

trying this application doesn't require

me to change how I currently manage my

calendar, but it's more or less a faster

way to get done what I already get done.

But the second statement is quite

different. Since it's an applications

that manage my entire calendar, I would

be more reluctant to try it because I

have to relinquish how I already manage

my calendar to this new agentic

application. So the core difference here

is that AI applications helps you solve

the problem you already know that you

have since when you ask something it

completes calendar tasks but within your

existing workflow that you might already

be satisfied with. But Agentic

application is quite different. It tries

to take over the entire workflow even if

you never felt like the workflow itself

was a problem. This kind of shift can

cause unnecessary friction. So there in

lies the current gap in AI adoption when

it comes to agentic application which is

are agentic applications truly solving

the right problem that people have.

What's actually common to see is that we

have a situation where agentic

applications become an overkill and

people essentially reduce agentics

application into a simple task

completion tool. Sort of like owning a

Lamborghini but living in a school zone.

And that's because while an agentic

application might be more powerful than

AI application, solving problems that

exist on a workflow level is so unique

and complex that it's hard to not throw

the baby with a bathwater. And here's a

really cool analogy. In computer

science, there's a common metric called

time and space complexity where it

measures the scalability of performance

as the input size scales. For example,

if the complexity is O of N^ squ, that

means that if you double the input size,

the upper bound in time and space that's

needed to solve that problem is

quadrupled. So this makes scaling very

inefficient. Now, if you transfer this

kind of framework into AI adoption and

the input size is human resources, we

can draw a similar insight where as the

size of the organization grows, the

complexity in adopting AI likely grows

unscalably. In other words, adopting AI

into an organization that has 100

employees will take proportionately less

complexity than an organization that has

a thousand employees. And this is why

most gains in value from AI are still

reduced down to the individual rather

than the sum. In other words, employees

are finding ways to improve their

individual tasks to code faster, code

better, and execute tasks rather than

trying to use AI on an organizational

level. And trying to force AI into an

organization often creates an uglier

outcome at the expense of making some

parts really good. And this is also

reflected in the Dora report that says

that AI is an amplifier that magnifies

the strength of high performing

organizations and the dysfunctions of

struggling ones. The report goes on to

say in well-aligned organizations AI

amplifies flows in fragmented ones it

exposes pain points. So there in lies

the question is AI adoption capped at

the task level? Meaning, are AI

applications only as helpful as helping

individuals complete tasks? Or are we

yet to see how AI can transform beyond

individuals in completing tasks and

towards a true agentic system where it

can autonomously decide how to interact

with the real world and bring value in

an organizational level. Similar

sentiment can also be found in the Dora

report that says without intentional

changes in workflows, roles, governance,

and cultural expectation, AI tools are

likely to remain isolated boosts in an

otherwise unchanged system, a missed

opportunity. To scale AI's impact,

organizations should invest in

redesigning their systems. So, are we

better off adopting AI from inside out

rather than outside in? Meaning, will

organizations that master every tasks

with AI outperform newer organizations

that might adopt AI natively, but with

the advantage of pivoting to changing

demands by employing AI at every stage

of the company.