Latest / Elon Musk Podcast / 120 People Fired due to AI

Transcript

120 People Fired due to AI · Elon Musk Podcast

0:00
An unnamed e-commerce company eliminated its entire 120 person
0:04
copywriting team in a single stroke, replacing them with a
0:08
proprietary large language model to save $11 million annually.
0:12
Yeah, the the sheer brutality of that single decision perfectly
0:16
mirrors this broader, very quiet restructuring that is happening
0:20
across the global labor market right now.
0:22
Because if you look at the raw reporting, you know there have
0:24
been nearly 80,000 tech sector job cuts and almost half of
0:29
those eliminations are explicitly attributed to
0:32
artificial intelligence automation.
0:34
Rather than standard economic downsizing, we are looking at a
0:38
fundamental rewiring of corporate structure, and it is
0:41
happening in real time, right in front of us.
0:43
Right. And the headlines focus
0:44
relentlessly on those highly visible tech layoffs, right?
0:47
Which creates this illusion for you, listening to this, that
0:50
this is a highly localized phenomenon.
0:52
Yeah, exactly. Like it is contained entirely
0:55
within coastal software companies, but the reality is
0:59
there is this massive disconnect between those publicized tech
1:02
reductions and the, you know, systemic restructuring happening
1:06
across the non tech global workforce.
1:09
Right, like manufacturing finance.
1:10
Logistics. Traditional administrative hubs.
1:12
It is everywhere. Because when a major search
1:15
engine provider cuts 12,000 employees, I mean that makes the
1:18
front page of every financial publication, of course.
1:21
But when an industrial supply chain company in the Midwest
1:24
just quietly stops replacing its retiring administrative staff
1:28
because an autonomous system now handles all their procurement
1:31
workflows. Right, nobody raised an article
1:33
about that. Exactly.
1:34
It flies completely under the radar.
1:36
But that silent contraction effects far more people than the
1:39
dramatic corporate announcements.
1:42
The invisible shrinking of the workforce is where the true
1:44
economic impact lies. So with autonomous systems
1:48
stepping into complex human roles like this, how can we
1:51
accurately distinguish the workers who will be augmented by
1:55
this technology from the ones who will simply be substituted?
1:58
That is the core issue and the macroeconomic data paints a
2:02
really revealing picture regarding global hiring activity
2:06
right now, OK. Right now, global recruitment
2:08
remains roughly 20% below previous economic benchmarks.
2:12
20%, that is huge. It is.
2:14
And when you look at advanced economies, that contraction is
2:17
even more severe. They are frequently hitting
2:20
drops of up to 35% compared to historical norms.
2:24
The doors to new employment are heavily guarded right now.
2:28
And, you know, the natural public assumption is that
2:30
automation is out there actively destroying all these jobs, which
2:33
is creating this massive employment deficit.
2:36
But data from global professional networks indicates
2:39
something entirely different. It is restrictive monetary
2:42
policies and the high cost of capital that are the actual
2:45
primary drivers suppressing recruitment, right?
2:48
The net present value of a new hire has basically plummeted.
2:51
Yeah. We really have to look at the
2:53
mechanics of corporate finance to understand the gravity of
2:56
that plummet, OK? Lay it out for us.
2:58
Think of hiring an entry level employee like purchasing a solar
3:03
panel for your home. A solar panel, OK, right?
3:06
When interest rates are at 0, money is essentially free to
3:09
borrow. You do not mind waiting 5 or
3:12
even 7 years for that solar panel to pay for itself in
3:15
energy savings. The cost of waiting is 0, right?
3:18
Because you aren't paying any interest on the loan.
3:20
Exactly. And for years, companies
3:22
operated in exactly that zero interest rate environment.
3:25
Businesses invested heavily in future growth, and that meant
3:29
aggressively hiring human capital.
3:31
Yeah, they were just scooping up talent.
3:33
They could easily afford to hire an employee whose financial
3:36
return to the business might not materialize for 18 or 24 months.
3:40
The discount rate applied to their future productivity was
3:43
negligible. Wait, back up.
3:46
If the high cost of borrowing money is what is keeping hiring
3:49
down, why are so many companies explicitly pointing the finger
3:53
at automation in their layoff announcements?
3:56
Because it comes down to a dynamic economist call the low
3:59
higher, low fire environment. Low, higher, low fire.
4:02
Yeah, companies are terrified of losing their most experienced,
4:06
productive talent because replacing senior staff is
4:08
incredibly expensive and risky. So they hoard their existing
4:12
senior talent. But simultaneously they put up a
4:15
massive wall against new entrance.
4:18
They use automation not necessarily to fire the senior
4:21
people, but to entirely avoid hiring junior people.
4:24
Oh, I see. And when a company does announce
4:27
A layoff and publicly blames automation, it is a highly
4:31
convenient narrative. Because it sounds innovative.
4:33
Right. It pleases investors who want to
4:35
see the company embracing efficiency.
4:38
That narrative masks the harsh reality that the company simply
4:42
cannot afford its previous bloat due to rising debt costs.
4:47
This restrictive environment severely limits traditional
4:49
corporate career paths. I mean, it creates A brutal
4:52
bottleneck for junior candidates trying to get their foot in the
4:54
door. Yeah, it really does.
4:56
Consequently, this opens up a massive surge in professionals
4:58
turning toward entrepreneurship, independent creator roles, and,
5:02
you know, skill based technical trades.
5:04
People are realizing the corporate ladder is just missing
5:07
its bottom rungs entirely. The data shows significant
5:10
spikes in individuals registering as founders or
5:13
moving into specialized trades. If the corporate entry level
5:17
path is blocked by a combination of high interest rates and entry
5:20
level automation, talent naturally flows elsewhere.
5:23
Makes sense. The traditional safe bet of a
5:25
corporate job is vanishing, forcing an entire generation to
5:29
redefine economic security. So to really understand this
5:33
massive disconnect between the macro data and corporate claims,
5:38
we have to look at something called the iceberg index.
5:40
Yes, this is fascinating. It was developed by researchers
5:43
using a massive supercomputer to simulate millions of workers and
5:48
thousands of skills. The methodology behind the
5:50
iceberg index is just incredible because instead of relying on
5:53
past employment statistics, researchers built a large
5:56
population model. Like a simulation.
5:58
Exactly. Think of it as a highly detailed
6:01
digital twin of the entire economy.
6:03
They simulated 151 million individual workers. 151 million.
6:09
Yeah, and they mapped those workers across 32,000 distinct
6:13
skills and 3000 geographic counties.
6:16
Then they introduced thousands of digital tools into the
6:19
simulation to observe the interactions at a really
6:22
granular level. So like, take someone we will
6:25
call Sarah, right, a supply chain coordinator in Ohio.
6:28
OK, her job title does not sound highly technical, but the
6:32
supercomputer did not look at her job title.
6:34
It looked at her specific daily actions.
6:37
Right, the actual tasks. Exactly.
6:39
It identified that 40% of her day involves moving data from an
6:43
emailed shipping manifest into a logistics database.
6:46
Yeah, and that is a task a digital tool can execute in
6:49
fractions of a second. So the simulation mapped exactly
6:52
how vulnerable those specific skills are.
6:54
And the iceberg metaphor perfectly illustrates the
6:56
findings. The surface index, so the part
6:58
of the iceberg you can see clearly above the water that
7:01
represents the highly visible tech.
7:03
Hubs. Software engineers in coastal
7:04
cities. Right.
7:06
And this visible disruption comprises only about 2% of total
7:09
wage value. But the massive hidden block of
7:13
the iceberg submerged beneath the surface tells a completely
7:16
different story. Because that hidden mass
7:18
represents nearly 12% of the labor market, we are talking
7:21
about over a trillion dollars in wages and it is completely
7:25
geographically distributed. It comprises administrative,
7:29
financial and professional service roles scattered
7:32
nationwide in places like Ohio and Tennessee.
7:35
Exactly. These are the supply chain
7:37
coordinators, the compliance officers, the back office
7:40
financial analysts. The jobs being automated aren't
7:43
on assembly lines anymore, they are sitting in office cubicles.
7:46
Yeah, That is such a simple way to put it and it highlights a
7:49
critical flaw in our traditional economic indicators.
7:53
I mean, metrics like gross domestic product and
7:55
unemployment rates completely failed to capture this exposure
7:58
because they only measure outcomes after the disruption
8:01
happened. They are.
8:01
Lagging indicators. Right.
8:03
An unemployment rate only ticks up after someone has already
8:06
lost their livelihood and filed for benefits.
8:09
It measures the damage, not the vulnerability.
8:12
Because traditional metrics completely ignore the latent
8:14
overlap between human skills and machine capabilities.
8:17
Exactly. Going back to Sarah in Ohio, if
8:21
an autonomous system can suddenly do her data
8:23
reconciliation perfectly, she is highly exposed, but the
8:27
unemployment rate does not register that exposure until her
8:30
company actually decides to eliminate her role.
8:32
And the iceberg Index measures that precise technical exposure
8:36
before adoption crystallizes. It quantifies the wage value of
8:41
skills that systems can perform within each occupation.
8:44
Right. It proves that the capability
8:46
overlap is five times larger than the visible tech disruption
8:49
we see in the news. It's happening in white collar
8:52
hubs far removed from traditional tech centers.
8:55
Which is why relying on outdated metrics severely limits a
8:58
state's ability to protect its workforce.
9:01
If a state government waits for the unemployment rate to spike
9:03
before launching retraining programs, they are years too
9:06
late. They.
9:07
Are totally behind the curve. Yeah, this opens up the absolute
9:10
necessity for simulated skills based tracking.
9:14
Local governments must identify vulnerable communities before
9:17
mass displacement occurs. Local leaders need predictive
9:20
intelligence. They need to know that a
9:23
specific concentration of administrative roles in a
9:26
particular county has a high capability overlap with emerging
9:30
technology right now, right? Because that allows them to
9:33
direct infrastructure and reskilling investments
9:36
proactively. Which brings us to the concept
9:38
of replacement velocity. OK, what is that?
9:40
This is the speed at which partial task automation
9:43
transitions into full roll redundancy for jobs like
9:46
writers, programmers, and junior data analysts.
9:49
Because an occupation is basically just a bundle of task.
9:53
Like a junior data analyst cleans data, runs database
9:56
queries, formats reports and presents findings to management.
10:00
If a system automates the data cleaning, the analyst is
10:03
augmented. They suddenly have more time to
10:06
focus on the presentation. But if the system eventually
10:09
automates the queries in the formatting, the remaining
10:12
presentation task might not justify a full time salary.
10:16
The velocity measures how fast that exact tipping point is
10:19
reached. And we are witnessing the
10:22
historical inversion of Labor risk.
10:24
The historical inversion. Yeah, because previous
10:27
industrial revolutions targeted manual labor.
10:29
The mechanical loom replaced the physical labor of the Weaver.
10:33
The robotic arm replaced the repetitive motion of the
10:35
assembly line welder. Right, the physical tasks were
10:37
always the first to be mechanized.
10:39
Exactly, but this technological wave specifically targets high
10:43
wage cognitive heavy sectors. The historical pyramid of lower
10:47
risk has been flipped completely upside down.
10:50
The only true motes remaining are physical dexterity and high
10:53
level complex reasoning. See, I disagree entirely with
10:56
the premise that all desk jobs are equally threatened just
10:59
because they involve cognitive work.
11:01
I think it's severely misunderstands what makes
11:03
certain human roles valuable in a business context.
11:06
But if a system can process information and generate text
11:09
faster than any human, it inherently threatens the desk
11:13
worker whose entire job is processing information and
11:16
generating text. Well, it threatens the routine
11:18
processor, sure, but it's severely underestimates roles
11:21
requiring high emotional intelligence, nuance,
11:24
negotiation, and complex interpersonal judgement.
11:27
You think those are safe? I think they are highly
11:29
resistant to substitution. Look, a system can generate a
11:33
legally flawless merger contract in 10 seconds.
11:36
It cannot sit in a boardroom, read the defensive body language
11:39
of the opposing counsel, understand the unstated
11:42
financial fears of the client, and navigate the delicate ego
11:45
dynamics required to actually get signatures on that contract.
11:49
You are drawing a very hard line between cognitive synthesis and
11:52
emotional application. Because synthesizing data is
11:55
vulnerable, applying data through human trust is highly
11:58
protected. Trust cannot be computed.
12:01
Leadership, empathy, conflict, mitigation.
12:04
These are fundamentally human to human protocols.
12:06
That's fair. A machine can provide the
12:08
optimal negotiation strategy. Executing that strategy requires
12:12
a human to convince another human to take a professional
12:15
risk. And this dynamic drives AK
12:17
shaped labor market right? The top arm of the K shoots
12:20
upward, representing highly skilled specialists who combine
12:24
deep domain expertise with the ability to leverage these new
12:27
tools. They command massive premium
12:30
salaries because their output is multiplied exponentially.
12:33
They become super workers. Exactly.
12:35
But the bottom arm of the K points downward, representing
12:38
generalists who only perform routine cognitive tasks.
12:41
They face stagnant wages and shrinking opportunities because
12:45
their court output is now just a cheap commodity.
12:48
In the middle basically hollows out.
12:50
You are either directing the systems or you are competing
12:53
against them. The premium is placed on
12:55
judgement over execution. Yes, if execution is essentially
12:58
free, the market rewards the person who knows exactly what
13:02
needs to be executed to achieve the business goal.
13:04
Which drastically limits the value of a generalized college
13:07
degree. For decades, a generic bachelors
13:10
degree was a ticket to the middle class because it signaled
13:13
A baseline level of literacy and cognitive compliance to an
13:16
employer. Right, it was a filter.
13:18
But that is no longer enough. We are entering a reality where
13:22
specific, demonstrable skills dictate economic survival.
13:26
An employer does not care about a piece of paper.
13:29
They care about your verifiable ability to solve a complex
13:32
problem or navigate a difficult human interaction.
13:35
Which makes sense when you look at the transition from simple
13:38
chat bots to autonomous systems. I mean, we are moving beyond
13:41
systems where you type a question and get a static
13:43
answer. We are looking at frameworks
13:45
like Google, Sage, Anthropics, Claude Cowork, and Open AI is
13:50
Operator. Yes, this is the shift from
13:52
instruction based to intent based computing.
13:54
OK. Explain the difference.
13:56
In the old instruction based model you had to provide step by
13:59
step commands. Open the spreadsheet program,
14:02
click Cell before run the summation formula.
14:04
Save the document as a PDF. Open the e-mail client attached
14:08
to the file and send it to the finance department.
14:10
Right, you had to hand hold the computer.
14:12
Exactly. The human is the orchestrator of
14:14
a dozen different micro actions. But in intent based computing,
14:19
you define a desired outcome. You just tell the system,
14:22
reconcile last month's vendor invoices against the shipping
14:25
logs and notify anyone with a discrepancy, and the system
14:29
autonomously plans, coordinates, and executes the necessary
14:33
workflow across multiple applications.
14:36
It opens the database, runs the comparison, drafts the
14:39
individual emails, and sends them.
14:41
The human never manages the intermediate steps.
14:44
But hold on, how do we know these systems aren't just parlor
14:47
tricks dressed up with good marketing?
14:49
Because we have all seen highly edited demonstration videos from
14:53
tech companies that completely fail to work in the real world.
14:56
That is a great question, and a highly detailed technical guide
14:59
recently exposed the exact difference between fake agents
15:02
and real ones. A fake AI agent is like a train
15:04
on a track. It looks like it is moving
15:06
incredibly fast on its own, but if there is a fallen tree on the
15:09
tracks, the train simply crashes.
15:12
It is just a series of sequential API following a rigid
15:15
script. If a database is down or an
15:18
invoice is formatted differently, the fake agent
15:20
completely shatters because it has no capacity to reason
15:24
through the obstacle. Ah I see.
15:25
Whereas a true intent based agent is like an off road
15:29
vehicle equipped with the GPS. Exactly.
15:31
If the bridge is out, it automatically stops, backs up,
15:34
calculates a new route, and finds a dirt road to get to the
15:37
destination. True production ready agents use
15:40
complex parallel and loop execution patterns.
15:43
They can pivot. Yeah, they try a solution,
15:45
evaluate the result, realize it failed, and autonomously try a
15:49
different approach. They possess A rudimentary form
15:52
of trial and error. But this level of autonomy
15:54
requires strict monitoring infrastructure to prevent
15:57
disasters. I mean the technical guide sites
16:00
a recent incident where a company deployed an agent
16:03
without proper guardrails. Oh, the story is wild.
16:06
Right, the system encountered a firewall error while trying to
16:08
access a cloud service. It attempted to resolve the
16:11
error, failed, and got stuck in an unmonitored runaway loop.
16:14
It just kept trying. Yes, it repeatedly called the
16:18
external software service, spinning up new instances
16:22
multiple times a second. That single loop generated
16:25
$47,000 in software usage fees before a human engineer noticed
16:30
and pulled the plug. That perfectly illustrates the
16:32
power and the danger. And this shift completely
16:36
changes the software industry. It severely limits the dominance
16:39
of traditional software as a service models.
16:42
How so? Well, for the last decade,
16:44
companies paid per seat licenses for static software a business,
16:48
but 100 seats of a customer relationship management tool for
16:51
its 100 sales representatives. Right, but if an intent based
16:55
agent can autonomously fetch data, update records and
16:58
generate reports across your custom databases, you do not
17:01
need to pay for a massive bloated software platform.
17:04
You just need the agent. Exactly.
17:06
This realization caused massive market panic.
17:08
It wiped hundreds of billions in market value from traditional
17:11
software companies as investors realized enterprises might build
17:14
custom autonomous workflows instead of paying endless
17:17
subscription fees for static tools.
17:19
Which? Introduces the economic concept
17:22
of demand expandability. Task automation does not
17:25
automatically equal job loss. We can look at this through the
17:28
lens of the Jevons paradox. The Jevons paradox, Yeah.
17:32
William Stanley Jevons observed that when technological
17:34
improvements increased the efficiency of coal use, the
17:37
overall consumption of coal skyrocketed.
17:41
Because it became cheaper and more efficient to use coal
17:44
society found entirely new uses for it.
17:46
Right. So when the cost of a resource
17:48
falls, consumption of that resource generally rises.
17:51
And applying the Jevons Paradox to cognitive labor alters
17:55
everything. If you make a worker twice as
17:58
efficient, you do not automatically fire half your
18:00
workforce if the demand for their output is expandable.
18:03
You hire more people because the cost of producing that output is
18:06
dropped, generating entirely new markets.
18:08
You can really see this if you contrast a call center
18:11
representative with a software engineer.
18:14
A call center generally has a fixed volume of customer
18:16
inquiries. A business does not suddenly
18:18
want more customer complaints just because they can process
18:21
them faster. Obviously not.
18:22
The demand is strictly capped, so making a call center highly
18:26
efficient through automation leads directly to substitution
18:30
and job loss. The required human headcount
18:33
shrinks. But an enterprise has an
18:35
infinite latent demand for digital products.
18:38
There's always another feature to build, another internal tool
18:41
to optimize, another workflow, to digitize.
18:45
Making coding cheaper and faster does not mean a company needs
18:48
fewer software engineers. It causes the enterprise to
18:51
build far more software. The output expands
18:54
exponentially. That requires humans to direct,
18:57
manage, and secure that rapidly expanding output.
19:00
So making a process highly efficient does not destroy the
19:03
profession if the world has an endless appetite for the final
19:06
product. That is exactly it, and that
19:08
dynamic limits the accuracy of standard displacement
19:10
predictions. When analysts look at a
19:12
spreadsheet and calculate that autonomous systems can write
19:15
code 40% faster, they immediately subtract 40% of
19:18
engineering jobs. But that assumes the total
19:21
amount of code needed in the world is a fixed number.
19:24
It ignores demand expandability completely.
19:27
This opens up massive growth potential for industries with
19:30
uncapped consumer demand. Entertainment, Personalized
19:34
healthcare, custom software development.
19:37
Drastically lowering the cost of production in these fields means
19:40
the market will aggressively consume the increased output.
19:43
And we are already seeing this because despite the overall
19:47
hiring slowdown, over 1,000,000 new technology enabled roles
19:52
have been created recently. Yeah, the new collar workforce.
19:55
Exactly. We are seeing the rise of these
19:58
new collar workers. These are roles like data
20:00
annotators and forward deployed engineers.
20:03
They are not traditional blue collar manual labor and they're
20:06
not traditional white collar management, right?
20:08
They are specialized technical positions focus entirely on
20:11
bridging the gap between autonomous systems and real
20:14
world implementation. And the organization's
20:16
successfully integrating this new collar workforce are
20:19
mastering a metric known as talent.
20:21
Velocity, talent velocity, Yeah. This measures an organization's
20:24
ability to clearly see it's available internal skills,
20:28
rapidly acquire what is missing, and mobilize talent continuously
20:32
to meet shifting market demands. Which most companies cannot do.
20:35
The vast majority of companies completely lack this agility.
20:38
They are stuck in rigid, decades old human resource paradigms.
20:43
Most companies still operate on static job descriptions.
20:46
You are hired as a level 2 Financial Analyst and you sit in
20:49
that exact box doing those specific tasks until you are
20:53
promoted to a Level 3 Financial Analyst.
20:56
It is slow, inflexible and entirely unsuited for an
20:59
environment where the nature of the tasks changes month to
21:02
month. But the most successful
21:04
organizations are abandoning those rigid job titles.
21:07
They are moving aggressively towards skills based hiring over
21:11
traditional credentialing. They do not care if you have a
21:14
specific university degree. They care if you have a
21:17
demonstrable proficiency in system architecture or conflict
21:20
mitigation. They maintain dynamic data basis
21:23
of their employees exact capabilities.
21:26
And instead of departments operating in silos, they form
21:28
fluid teams based on the specific capabilities required
21:31
for a project. If a new initiative requires
21:34
data structuring, ethical oversight, and front end design,
21:37
the organization pulls the individuals with those exact
21:40
skills into a temporary unit, regardless of their official job
21:44
titles. Once the project is done, the
21:47
team dissolves and the individuals flow into new
21:50
initiatives. This fundamentally changes
21:52
internal corporate structure. It's severely limits the
21:55
traditional hierarchical career ladder.
21:58
You do not climb straight up a single departmental silo
22:01
anymore. It opens up horizontal internal
22:04
mobility as the primary method for talent retention and skill
22:07
deployment. A worker might shift across 5
22:09
different departments over three years, continually applying
22:12
their core skills to entirely different business problems.
22:16
It requires a massive cultural shift regarding how we define a
22:18
successful career. Right?
22:20
So to sum this up, corporate output is completely decoupling
22:23
from traditional hiring practices.
22:25
The organizations and individuals thriving in this K
22:28
shaped economy are those who discard rigid job titles in
22:31
favor of fluid, intent driven skill application.
22:34
And it leaves you with this final thought.
22:37
If the most successful future organizations are essentially
22:40
human agent collaborative teams, what happens to the traditional
22:44
entry level ecosystem where professionals historically
22:47
learned the very judgment skills they now need to manage the
22:50
machines? If you're not subscribed yet,
22:53
take a second and hit follow on whatever app you're using.
22:56
It helps us keep making this. We appreciate you being here.