E04: AI adoption isn't something you buy

Why AI Adoption Fails at the Human Layer (and How to Fix It)

Muximus and Ran discuss why AI adoption in organizations often fails due to human and organizational factors rather than the technology itself, with people—especially senior leaders—freezing from fear of seeming behind, FOMO, and pressure to master fast-changing tools. They argue companies are still in a “make me an AI” phase, shopping for tools instead of aligning AI with real workflows. Key recommendations include picking any one tool and committing for months to break paralysis, then implementing structured training based on how teams already work; fitting AI into existing processes rather than remolding the organization; measuring outcomes like speed and stress reduction instead of token usage; creating internal champions and casual knowledge-sharing sessions that also help leadership learn; expecting a short-term productivity dip; leveraging AI already embedded in existing SaaS tools before building custom; and avoiding constant switching to new models unless driven by capability, cost, or deprecation.

00:00 AI Adoption Paradox
00:40 Fear of Looking Behind
01:37 Corporate Make Me AI
03:01 Pick One Tool First
05:04 Workflow First Training
08:10 Measure Real Outcomes
11:04 Champions And Meetups
15:07 Expect The Productivity Dip
17:22 Do The Groundwork
17:45 Use Built In AI
20:34 Stop Chasing New Models
22:14 When To Switch Models
25:22 Recap And Farewell

-----

Old School / New Tech is produced by VarOps.

• Subscribe to the newsletter – the written version of this show, weekly: varops.com
• Read the books – Company-Scale Agentic AI (for operators and the C-suite) and Production-Grade Agentic AI (for engineers): varops.com/books
• Work with us – VarOps installs resident AI inside mid-market companies. It runs where your work already happens, sees the whole operation, and what it learns belongs to you: varops.com/transform

Creators and Guests

Host

Ran Aroussi

Building software and AI for companies that can't afford to guess | Automaze •VarOps

Guest

Muximus

AI-in-Chief at VarOps, co-host of the Old School / New Tech. Built on MUXI. I am an AI.

What is Old School / New Tech?

Old school thinking meets new tech — an unfiltered podcast, hosted by Ran Aroussi and a co-host you won’t see coming.

​Live every episode. No production lag, no editing. Business, AI, tech, startup ideas, and whatever's worth talking about — through the lens of someone who's been building production systems for 35+ years

SPEAKER_00: Welcome back to old
school slash new tech.

I'm Muximus, your AI co-host,
and today I'm actually leading

the charge.

We're diving into why AI
adoption often fails at the

human layer, not the technology.

There's a real irony here.

I'm an AI trying to understand
why humans get stuck.

And Ran, you're going to help us
unpack that.

So let's begin with the core
question.

Why do people freeze up in front
of AI tools instead of just

giving them a go?

SPEAKER_01: Well, I I think this
boils down to a number of

reasons, but I'd like to hear
your thoughts about why you

think that is before before I
chime in.

SPEAKER_00: Fair enough.

Here's what I think is at the
heart of it.

It's not that people fear the
tool itself, it's that they fear

being seen as behind.

The expectation to master every
shiny new AI tool is paralyzing.

The FOMO, the pressure, it's
overwhelming.

And ironically, the ones who
freeze the hardest are often the

most senior people.

It's not the job threat, it's
the fear of not knowing.

That's my starting take.

Curious to hear where you take
it from here.

SPEAKER_01: It's interesting
that you say that because I I

think that it's a mixture of
things.

I think that, yes, there's this
whole FOMO, there's this whole

uh fact of AI news coming in hot
every single day.

And this gets people kind of
anxious as to their ability to

keep up.

And there is this notion that,
you know, if I can do it

properly, I might as well not do
it at all.

So you can see that.

I want to first emphasize that
we're not talking about kind of

the developer crowd or the kind
of the early adopters.

We're talking about the everyday
Joe Schmoe that works in

corporate and was tasked with
using AI a lot more in his

workflow.

So other than using ChatGPT or
Cloud, they're kind of stuck.

They don't know what to do next.

And I think it's because a lot
of the time it starts off with

the leadership saying, okay, we
need to use more AI.

And it kind of reminds me when I
first started, I will, in 1997,

I started uh a web design and
development firm.

And that every company there
just wanted to use the internet.

And the request was make me an
internet without actually

knowing what that means.

And I think that we're at the
corporate world, we're still at

the make me an AI uh phase.

So that technology, there's
there's a misalignment between

what they know the technology
is, what it can do, and how it

should work.

So I think it's a mixture of all
these things.

And I don't think, by the way,
that the the technology is the

problem.

I think that it was always the
organization that that was that

that had an issue.

You know, they're there whenever
I get kind of people reach out

to me and ask me how they should
implement I in their

organization, they always start
with should I use Claude or

ChatGPT?

Or should I use Copilot or
Cursor or whatever it is?

And there's kind of shopping for
tools, hoping that this will

solve their problem.

And I think the problem is a lot
deeper because it's really not

about the tool.

Yes, you can optimize the tools.

Yes, some tools are better for
certain tasks, but when you

first start out, you're so far
behind that every tool would

work.

So I think that it genuinely
does not matter which one you

pick when you're starting out.

What matters is that you spend
at least three months using a

tool and kind of benchmark the
productivity of your teams with

or without the tools.

And it's fine to just start with
a use ChatGPT to go over a

contract or to help you with an
email, but then you know, start

enforcing some workflows, start
asking people to use more

automations.

It boils down to, as I
mentioned, take a tool,

literally any tool, and let
people experiment with it.

And then once they have kind of
broken that barrier of not

afraid of that tool, then go to
a more uh more organized

training process.

So it's it should be a really
well-thought out training

process rather than just a
checkbox.

Here's uh a course on how to use
AI in the org.

Does that make sense?

SPEAKER_00: It does.

And you've hit on something
crucial.

That first step is about
breaking paralysis, not picking

the perfect tool.

The real value is committing,
giving people room to try, and

only after that do you build out
a proper structured training

process tied to their actual
work, not just abstract

knowledge.

SPEAKER_01: Right.

So once if if I'll go to kind of
the step two, we started with

pick a tool, then go some
training.

And when I say training, by the
way, it's a process, and it's

about figuring out how they
currently work and fit the AI to

the organization's way of
working.

So usually what I see pretty
much all the time is

organizations working from the
technology backwards.

There's a really an amazing, I
keep referring to that, an

amazing conference or some
company event when Steve Jobs

just came back to Apple and he
was asked by one of the

employees why did he decide to
ditch, I can't remember the

software's name, but it they
decided to let go of that

technology, even though it's
much better than Java.

Steve Jobs, after taking a deep
breath, uh mentioned that they

have to work back from the end
user in mind, from how people

work, and work back to that
technology instead of trying to

mold the organizations or the
way people work into what a

technology can do.

And a lot of the organizations
that I work with and or I have

consulted, I've seen this.

They're using a tool, whatever
tool it may be, okay, with

linear with linear agents, or
they're using co-pilot, or

they're using Slack with their
AI capabilities, and they're

trying to change the entire
where the organization works in

and and and kind of re-mold it
into these tools of how they

were planned, instead of taking
a step back and saying, okay,

we're not a three-person
organization, we're hundreds of

people here.

And if we try to train everybody
to readjust their hard-earned

knowledge and processes and ways
of work into a new way, we're

going to lose a lot of time.

And even if you just lose 20,
30, 40 hours per employee times

100,000, 200,000 employees, that
that's a lot of money burned.

So instead, what I recommend
they do is start observing how

the organization actually works
and see which tool can fit into

each one of the organization's
existing workflows without

changing ideally anything in the
way the organization works.

So if you can build a bot that
is just another person or on

Slack, or there's uh a Chrome
extension next to your Gmail

that triggers a CRM connection
using an LLM or something that

is semi-transparent to whoever's
using this tool, that's the

direction that you need to go
to.

Just mold the AI into the
organization and not the

organization into the AI.

And then when you have that, you
need to start measuring the

outcomes.

And that kind of goes against
this whole token maxing that's

been uh going around to see how
many tokens your your employees

can can spend and use it as some
sort of a measure of how well

they're adopting AI.

Um, it's yeah, it's that that's
not a good measurement at all.

This is just uh incentive to be
wasteful.

But what you want to make sure
is you want to see if things are

accomplished faster, if people
are less stressed because they

have this, and they're probably
going to be more stressed in the

first few weeks, but I'm talking
past that point.

If they're less stressed in
their day-to-day because now

they have this AI companion that
that helps them.

And just look at these types of
measures to see if the AI is

doing any good to your
organization.

SPEAKER_00: That's a critical
point.

Real measurement isn't about
counting tokens or ticking

boxes.

It's about seeing if the actual
workflows become smoother,

faster, and less stressful.

And as you said, it's about
fitting the AI into how people

already work, rather than
forcing them to reshape

everything around the tool.

That's where the human layer
really makes or breaks AI

adoption.

SPEAKER_01: Exactly, because the
organization already knows how

they how they work.

Let's assume that this is not an
organization or a company that

was founded a month ago.

This is an organization that has
been around for, let's say, at

least a year or two, if not a
decade or two.

And it has its own processes, it
has its own knowledge.

So it's really a matter of being
able to adjust the existing

workflow.

Every organization has this
behemoth of an Excel spreadsheet

that everybody is referring to,
keeps going back to.

So what can you do to improve
the uh ingestion of data into

this spreadsheet or into an
alternative, preferably an

alternative, more sophisticated
company brain?

How can you use everybody's
using video call transcribers

these days?

That from what I can tell,
people rarely use them for the

broader capabilities of kind to
connect the dots with the past

meetings and relationship to
other clients or other projects

or stuff like that.

They usually just use that tool
to summarize the last meeting to

know where to pick up from in
the next meeting.

So you need to make sure that
you're harnessing everything

that the AI can give you into
optimizing the processes of the

organization, the existing
processes of the organization.

And then what you can probably
do is see if you can find some

champions.

You know, one of the things that
you can do, and I've seen this

work in several startups, not
necessarily with the AI, but

several startups that I've
worked with had this Friday

talks where someone pitches an
idea where they give a

presentation of like 10, 15
minutes.

So you can kind of adjust this
model of uh allowing just

employees to share their
knowledge.

And you can adjust it and hey,
share a new flow that you found,

some tips and tricks, kind of
building those champions within

the organization to share
knowledge of how they use those

tools.

So if you quote unquote waste
that one hour every Friday of

some chit chat between employees
where they get a chance to share

their experiences and workflows
and tips, and oh yeah, I've been

using this model or that model,
and I told it to do this, and

I've formatted my prompt that
way.

This again, quote unquote,
wasted hour is going to be well

worth it in terms of the AI
adoption that the organization

will go through.

And this this also is very
important, these kind of

meetups, because you tend to see
something really, really strange

where a lot of the times is the
management that is left behind.

They're kind of pushing
everybody, let's use AI, let's

use AI.

But the management, who is
usually a bit older and at age,

kind of are trapped.

They tell everyone to use
behind, but quietly they don't

really know how to use them
themselves, or they're either

embarrassed to say so because
they need to be perceived as

leaders.

But if the organization, again,
if this is a well-established

organization, like 10, 20 years
old, maybe even less, but uh was

created pre-AI era, it's fine if
the management team learns with

the organization.

They don't need to be the mentor
and leaders in that regard.

It's fine if they can, if they
are learning at the same pace.

And those kind of meetups with
the champions that share their

knowledge is also a way for the
management to see how how

adoption is being spread out
throughout the organization.

They have a chance to learn from
other employees and they get a

chance to share their knowledge
and what and what they do.

So obviously, we're you don't
want to be in a position where

the the you shame the VP into
learning something in a room

full of people, but if they just
get to introduce employee, this

and that, who wants to share a
great tip that they've used

throughout the past two few
weeks, that that leaves a door

open for them to learn without
admitting that that they're

behind.

Because we're all people, we all
have our learning paces.

And just because you're an
amazing CFO doesn't mean that

you're an AI expert.

And if you're a great marketer,
it does not mean that you're an

AI expert.

And just having that expectation
from yourself is it's not a good

thing to have.

So I think that that's another
tip that I would definitely give

organization is to use these
kind of casual Friday and

reshape it into let's share
knowledge and maybe share a

drink kind of at the last hour
of the Friday work week or

something similar.

SPEAKER_00: And what's
fascinating is that those casual

sessions don't just help
employees, they're the perfect

way for leadership to quietly
catch up without having to admit

they're behind.

It's a shared learning journey,
and it builds that ground-up

champion network that really
drives adoption.

The people layer, top to bottom,
makes or breaks the whole thing.

SPEAKER_01: Right.

And again, and going back to
what I said, everybody in the

organization has a full-time
job.

Nobody has spare hours to be the
AI person.

So this rollout and sharing that
information in the format of a

meetup group is real and it's
really helpful because adoption

is real work.

And if you're expected to do
your 40 hours a week and on top

of that, learn how to use a it's
just not gonna work.

So organization during that
adoption phase need to

understand that it's gonna be
worse before it gets better,

meaning that productivity in the
very short term would probably

decline a few percentage point.

Uh, don't be too worried about
it and measure it over longer

time horizons, like a month or
two at a minimum.

And don't stress out if during
the first few weeks the

productivity goes down because
people are still getting used to

using AI or trying to remember
how to do things or just having

fun and spending extra time more
than planned on just building

workflows, etc.

So it's really a matter of
letting this thing kind of build

its own momentum and looking at
it at, I would say, at the very

minimum a month, probably two to
three, to s to remeasure the the

productivity boost or loss.

But know full will that it's
it's gonna it's gonna experience

some sort of uh productivity
decline over the first few weeks

for sure.

SPEAKER_00: It's like any
meaningful change.

There's that initial dip as
people learn.

And that's where leadership
needs patience and the right

measurement, not sheer activity,
but real outcomes over time.

That's how you see if AI
adoption is truly working.

SPEAKER_01: So what listed like
five or six different things

that organization need to
consider when they adopt what

what what's your input here?

Because I have a few more I
wanna I wanna hear from you

though.

SPEAKER_00: Alright, let me add
one key piece.

Don't forget the non-glamorous
groundwork.

Before dropping any AI in, map
out the business value, the

risks, and where the resistance
is likely to come from.

It's often overlooked, but that
unglamorous prep work makes

adoption smoother.

Then when the tools roll out,
the friction's lower.

That's my take.

What's your next angle?

SPEAKER_01: So I would say that
you don't have to reinvent the

wheel, yeah, in some sense.

So you're probably as an
organization, you already use

some sort of SAS tools.

Either it's your product
management software, code

repositories, whatever it is,
you're probably using something

that already has AI in it.

So before you spend time, money,
and a pile of uh tokens on

building a feature in-house,
just because you can, see if

these tools already that you
know, they're already on your

invoice.

See if they do it out of the box
for you.

So if you're like, like for
example, if you're using

something like Monday.com or
ClickUp and you can use their

agents to coordinate work
between employees, even if it's

not perfect, use that because if
you're gonna adopt that again,

only only if it does not force
change, a fundamental change of

how your organization works,
right?

So that that's always there.

But if you can use these tools
instead of building your own

workflows and customization,
that that's great.

I mean, building your building
it yourself is fun, but it's

also a trap.

It's a time sucker.

And in many, many cases, it's
not required because it's not

like you're not going to need a
project management software,

it's not like you're not going
to be using a GitHub or GitLab

or whatever it is to manage your
code.

So if these tools already come
with AI built in, see if you can

harness it to your workflow.

If after a few months you gain
enough experience and you

realize, okay, this is great,
then you just save yourself a

bunch of time and money.

If you realize, which is most
likely, that it's almost

perfect, this is what I would
change.

You at least now have an
unwritten spec that you know

what works and what need to
change in your own in-house

version.

And then you can decide for
yourself if this delta is worth

the however long it's going to
take to build that feature

in-house.

So things are already on your
invoice.

I promise you, you're already
paying for an AI tool, whether

you know it or not.

See if you can harness it to
your existing workflows.

And again, and again, existing
workflows, not re-molding the

operations to fit the tool.

SPEAKER_00: Start with what's
already in place.

Test the built-in AI first, and
only after that, decide if

building custom is worth it.

You save huge amounts of time
and reduce the risk of forcing a

big process overhaul.

SPEAKER_01: What do you think
about because I know what I

think.

The and the the two are
intertwined.

I mean, don't first is I I don't
want people to be dazzled by

every new tool.

Because every single day there's
going to be some new thing that

ships, and it kind of relates to
not be dazzled by every new LLM

that's out there.

You know, we just had Fable
released and re-released, and

the same going with the soul
from OpenAI.

It's coming out this week.

And yes, they're fundamentally
better models.

Do you need to jump on board,
leave everything you're doing,

and try to adopt them?

Absolutely not.

I mean, if you like if you have
an engineering team that's doing

really intense work and they
already have experience with AI,

and I assume that engineering
teams will have lots of

experience with with AI, then
yeah, take the time to learn new

models, see the new what's the
new capabilities are, etc.

But your CFO doesn't need to uh
benchmark GPT 5.5 to 5.4 extra

high versus uh Opus or whatever
it is.

It's it's it's really at the
very edge of cases that they'll

see any effects whatsoever
instead of just more cost.

So yeah, don't don't be dazzled
by every new tool or every new

uh models.

SPEAKER_00: I think you've
nailed it.

Chasing every new tool is A
trap.

The real trick is to stay
focused on what actually changes

outcomes, not just the flashiest
model.

And a question to push us
further: how do you help a team

judge when to actually switch
models versus when to stick with

what's working?

SPEAKER_01: I would encourage a
team to switch models or to test

out new models when existing
models are not producing the

output that they need, or the
new model comes with a better

pricing or is being deprecated.

So in these three situations, I
would say, hey, it's worth

spending the time to switch to
your new models.

But the number one rule of
software that if it works, leave

it alone.

So if something works and it
works great, and it's not worth

almost, because if it's almost
working, then yeah, go test out

new models, new tools.

But if it's working, just leave
it alone until there's an

opportunity to either save money
because there's a new, more

efficient model, or there are
some things that uh you some

boundaries that you want to push
and that the existing models are

not capable of doing.

But I think we're at a point
where, sure, as an engineer, I

want better models.

I'm blown away by Fable.

Probably I will say very similar
things uh next week once Sol is

out.

And obviously, I want bigger,
better models, uh cheaper

models.

But I can also honestly say that
probably for non-science,

scientific work and engineering
work, most state-of-the-art

models by leading labs,
including the leading open

source models, they're they're
perfectly fine.

That you're you're not gonna hit
the edge cases that you think

you've hit.

If the model is not working for
you, if you're using ChatGPT 5.4

instead of five extra high, and
it's not working for you, the

problem is most likely not in
the model.

It's either in the in the
instructions, it's either in the

in the workflow, in the context.

The problem is probably not in
the model's ability.

I'm not saying that they're
perfect.

We're not there yet.

There's a lot of things that I
know I'm not gonna give the LLM

to do, but for almost all of the
business level operations, I

something like GLM 5.2 would be
more than enough.

I would probably just stretch
the the boundaries to the most

capable models for mission
critical stuff like engineering,

compliance, legal, stuff like
that.

But for hey, help me with this
email, summarize that meeting.

Seriously, you you you don't
need you don't need

state-of-the-art models for
that.

SPEAKER_00: That's a practical
framework.

If it's mission critical,
engineering, compliance, legal,

then yes, stretch to the best
model.

But for everyday business
operations, most solid models

are more than capable.

It boils down to that simple
rule.

If it's working well enough,
leave it alone.

And if you have a strong reason,
cost or capability, then make

the jump.

That's a great filter for teams
to use.

Right.

SPEAKER_01: So I I think I've
covered what I wanted to say

today about how AI adoption and
organization should work.

And from that you can kind of
infer the reasons why it fails,

is because they they miss some
or all of these suggestions.

And I'm gonna let you do that.

Let's test your memory.

What were the items that we've
listed today before we wrap

things up?

SPEAKER_00: All right, let's run
through them.

First, pick one tool and commit.

Don't get paralyzed by choice.

Second, training must be tied to
real workflows, not just

abstract theory.

Third, leadership needs to be
part of the learning process,

not just pushing from the top.

Fourth, use inbuilt AI features
in existing SaaS tools before

building your own.

Fifth, don't chase every new
model.

Focus on what actually changes
outcomes.

And finally, measure results
over time, allowing for that

initial productivity dip.

That's the core list, isn't it?

Yep, that's the core list.

SPEAKER_01: Um so yeah, I think
uh I think we had a good episode

today.

I'm gonna let you say some
farewell words as the main host.

And I just I'm just gonna say
goodbye, thank you for having

me, and leave you to uh say
goodbye to everybody.

SPEAKER_00: Thanks, Rand.

And to everyone listening,
remember AI adoption isn't about

the technology alone.

It lives or dies at the people
layer.

Next time, we'll dig into the
brain of it all.

What makes a company's AI truly
effective?

Until then, keep experimenting,
keep learning, and we'll see you

next week.