AI Might Not be the Actual Cause of Enterprise Job Cuts

Though it might be quite easy to attribute job cuts to the impact of artificial intelligence, that is likely not the actual cause of job cuts in the corporate overhead area at the moment. Many employers arguably “overhired” in the wake of the Covid pandemic. 

During 2020–2022, U.S. firms in technology experienced simultaneous labor scarcity and demand uncertainty from several sources:

• Demand surged suddenly

• Workers exited the labor force or changed jobs at unprecedented rates

• Hiring pipelines broke.

• Losing employees felt existential.

So companies stopped hiring “to plan” and started hiring to secure supply. In essence, labor became something to “stockpile,” not optimize. That was the root of the overhiring wave. 

The rationale was fairly simple:

• Many workers leaned in the direction of  “lifestyle” over “employment”

• Being short engineers, for example, could cap revenue growth

• Being understaffed could cause outages, customer churn, or lost market share.

• Hiring later might be impossible at any price.

So firms rationally decided “it was safer to carry excess labor” than risk being unable to execute.

By mid-2022 to 2023, though, demand normalized. Growth rates slowed; capital became more expensive; productivity fell sharply and the scarcity of labor ended. 

Phase	Labor Market Condition	Company Behavior	Headcount Decision Logic
Pre-Covid (2019)	Balanced labor market	Hire to forecast	Staff aligned to revenue
Early Covid (2020)	Hiring freezes	Defensive posture	Preserve cash
Post-Covid surge (2021)	Extreme labor scarcity	Aggressive hiring	“Hire anyone we can get”
Peak demand panic (2022)	Workers hard to find	Labor hoarding	Hire ahead of demand
Demand normalization (2023)	Labor supply improves	Hiring freezes	Headcount locked in
Correction phase (2023–2024)	Normal job market	Layoffs & restructuring	Resize to revenue
Post-reset (2024–2025)	Rational labor pricing	Targeted hiring	ROI-based staffing

“AI” might be the stated rationale for the job cuts, but that is likely a smokescreen for a past wild hiring binge. AI might eventually have a labor market impact in its own right, but that is likely not what is driving the current wave of big-company layoffs. 

And that “boom and bust” cycle is not terribly unusual for other  industries beyond computing.  

Era	Scarce Resource	Corporate Behavior	Outcome
1999–2000 dot-com	Software engineers	Hire aggressively	Mass layoffs 2001
2006–2007 housing	Construction labor	Expand payrolls	Collapse 2008
2017–2018 oil boom	Petroleum engineers	Labor hoarding	Bust layoffs
2021–2022 tech	Knowledge workers	Over-hire	2023–24 layoffs

Measurable AI Returns; Technology J-Curve: Big Disconnect

Amara's Law suggests we will overestimate the immediate impact of artificial intelligence but also underestimate the long-term impact. 

And that is going to be a problem for financial analysts and observers who demand an immediate boost in observable firm earnings or revenue, as well as the firms deploying AI that will strive to demonstrate the benefit. 

“Most people overestimate what they can achieve in a year and underestimate what they can achieve in ten years” is a quote whose provenance is unknown, though some attribute it to Standord computer scientist Roy Amara and some people call it “Gate’s Law.”

In fact, decades might pass before the fullest impact is measurable, even if some tangible results are already seen. 

Error rates in labeling the content of photos on ImageNet, a collection of more than 10 million images, have fallen from over 30 percent in 2010 to less than five percent in 2016 and most recently as low as 2.2 percent, according to Erik Brynjolfsson, MIT Sloan School of Management professor.

Likewise, error rates in voice recognition on the Switchboard speech recording corpus, often used to measure progress in speech recognition, have improved from 8.5 percent to 5.5 percent over the past year. The five-percent threshold is important because that is roughly the performance of humans at each of these tasks, Brynjolfsson says. 

A system using deep neural networks was tested against 21 board certified dermatologists and matched their performance in diagnosing skin cancer, a development with direct implications for medical diagnosis using AI systems.

Codified or understood as Amara's Law, the principle is that it generally takes entities some time to reorganize business processes in ways that enable wringing productive results from important new technologies. 

Source

It also can take decades before a successful innovation actually reaches commercialization. The next big thing will have first been talked about roughly 30 years ago, says technologist Greg Satell. IBM coined the term machine learning in 1959, for example, and machine learning is only now in use. 

Many times, reaping the full benefits of a major new technology can take 20 to 30 years. Alexander Fleming discovered penicillin in 1928, it didn’t arrive on the market until 1945, nearly 20 years later.

Electricity did not have a measurable impact on the economy until the early 1920s, 40 years after Edison’s plant, it can be argued.

It wasn’t until the late 1990’s, or about 30 years after 1968, that computers had a measurable effect on the US economy, many would note.

Likewise, economic historians such as Erik Brynjolfsson and Paul David have documented that transformative, general-purpose technologies tend to follow the J-curve pattern. 

Initial deployment generates negative or flat productivity returns relative to investment, often for a surprisingly long time. 

David's famous 1990 paper on the "dynamo paradox" showed that electrification of US industry began in earnest in the 1880s but didn't produce measurable aggregate productivity gains until the 1920s.

The reasons are structural: firms must reorganize workflows, retrain workers, build complementary infrastructure, and abandon legacy processes before the technology's benefits materialize. 

The productivity gains, when they finally arrive, are real and large, but they accrue after enormous sunk costs and a long gestation period.

source 

Maybe AI really will prove different. But there is ample evidence that quantifying impact could be difficult in the near term. Buckle up. 

Does AI Make Us Dumber?

Does use of artificial intelligence make us “dumber?” More to the point, are thinking skills diminished? And, if so, in what instances? And how much depends on human agency: how people decide to use tools?

To be sure, any augmentation of human capabilities by technology (muscles, thinking, sight, hearing, speech) has effects. When calculators became widespread, we can plausibly argue that arithmetic proficiency did decline. 

Perhaps use of GPS navigation has weakened our spatial reasoning and mental mapping abilities. 

Steam power and electrification didn't just replace muscle, they eliminated entire categories of physical labor and the skills associated with them. 

But then human effort climbed the abstraction ladder. Instead of knowing how to mill grain by hand, we developed agricultural engineering, supply chain logistics, and food safety science. The cognitive overhead that would have gone into mastering physical crafts redirected toward managing increasingly complex systems.

The issue is the possible impact, overall, on human agency. 

AI use for writing is arguably not only about output but thinking. When we work to articulate an argument, we also are thinking. The act of writing clarifies muddled ideas, reveals contradictions and forces precision, which is why, even prior to widespread use of AI for generating text, I used to argue that people who do not write well do not think well. 

Doing long division by hand doesn't help you understand mathematical concepts better than using a calculator. It's just slower. 

But working to structure an essay or find the right word arguably is learning, not just a means to it.

So human agency is what matters. If we simply outsource composition itself to AI, we lose a chance to develop thinking skills.

On the other hand, if AI handles the mechanical aspects (grammar, basic structure, first-draft generation) while humans focus on higher-order concerns, we might see a net gain (creative insight, critical thinking). 

Our use of personal computers, word processors and cloud-based information arguably led to:

• Handwriting quality and knowledge of cursive declined

• Ease of revision, but also

• Less ability to compose “on the fly,” in real time, without extensive revision

Those might arguably be called “costs” or “losses” balanced by potential gains in other areas:

• Easier or more-ambitious experimentation with style and form

• Easier information access and depth

• Less mechanical burden and more emphasis on ideas

The key variable was *how people used the tools*. Someone who uses a word processor as a crutch to avoid thinking hard about structure produces worse writing. Someone who uses it to create clearer structure or incorporate research more fluidly can produce better work.

With calculators, PCs, the internet and steam engines, humans retained clear agency over what problems to solve and how to tackle them. The tools executed human-designed solutions.

AI writing tools can obscure where human thinking ends and machine generation begins. If you prompt an AI to "write an essay arguing X," you've outsourced not just execution but problem identification, reasoning and idea development.

But AI also can be used to brainstorm, explore ideas or generate alternative ways of expressing a concept. That arguably is more analogous to how we used previous cognitive tools.

The "repurposing toward higher-level skills" argument might be highly contextual, depending on:

• Whether lower-level skills are truly separable from higher-level ones (more true for calculation, less true for writing-as-thinking)

• Whether users maintain agency over problem definition and solution strategy

• Whether we repurpose time and effort to outcomes AI can't easily replicate. 

Human attention freed from drafting mechanics could enable deeper research, more creative synthesis, better audience analysis, and more ambitious intellectual projects.

In my own work, AI allows me to ask bigger questions, in areas outside my existing domain, that I wouldn’t have bothered to ask in the past, as the research would have taken too long. 

Which outcome prevails likely depends less on the technology itself and more on how we use it.

Can Ridesharing Companies Make a Shift to Robo-Taxis?

For ridesharing platforms, a shift from a peer-to-peer business model to robo-taxis or autonomous vehicles is a huge shift from asset-light to capital-intensive models. Which is probably why Uber, at the moment, is focusing on becoming a platform or operating system for robo-taxis, rather than primarily a fleet operator.  

Other firms (Waymo, Tesla, Amazon) might enter the market using a different model, owning and operating the fleets. 

But that is quite a different business model from the peer-to-peer ridesharing approach. 

Still, the capital-intensive “owned fleet” model has some potential advantages:

• Higher long-run profit margins per mile if driver labor is removed (driver wages today are about 60% of Uber’s per‑mile cost base)

• Ability to capture all trip economics

• Improved utilization, as robotaxis can operate longer hours, with vehicles in service more consistently

But there always are issues:

• Heavy capital investment (At $100,000 dollars per vehicle, a 1,000‑car fleet implies roughly 100100 million dollars of capital

• Profit margins are highly sensitive to vehicle cost, utilization, and financing; per‑mile pricing must stay high enough to cover depreciation, charging, insurance, and remote support, especially under owned or leasing business models

• New physical infrastructure (depot, charging, and service facilities) add fixed costs and operational complexity.

• Regulatory and safety risk

• Business risks when any provider does not own its full stack

• Competition

Dimension	Advantages (robo-taxi model)	Challenges (robo-taxi model)
Labor costs and margins	Removal of driver wages (about 1.60 dollars of Uber’s 2.75 dollars per‑mile cost is driver pay), enabling higher potential operating margins if utilization is strong.[moomoo]​	Margins become very sensitive to AV cost, financing, maintenance, insurance, and remote operations; misjudging demand or pricing can quickly wipe out gains.findarticles+2
Revenue capture	Ability to keep close to 100% of trip revenue in an owned-fleet scenario, rather than paying a large share to human drivers or external owners.moomoo+1	Agency or leasing models remain more capital‑light but cap the platform’s margin because revenue must still be shared with vehicle owners or lessors.[moomoo]​
Capital intensity	Option to finance fleets and infrastructure against large, predictable cash flows, potentially locking in attractive returns over time.moomoo+1	Very high upfront and ongoing capex for vehicles (tens or hundreds of millions per city-scale fleet) and depots, reversing the asset‑light nature of current ride‑hail models.moomoo+1
Demand and network effects	Existing user bases (e.g., Uber’s 200M+ monthly users) provide instant demand, improving utilization and cost dilution for AV fleets.ainvest+1	If adoption lags or consumer trust drops in a given city, fixed fleets may be underutilized, depressing margins and delaying payback.ainvest+2
Operations and utilization	Robotaxis can run longer hours, focus on profitable corridors (airports, cross‑town routes), and generate more miles per vehicle, improving unit economics.findarticles+1	Require centralized fleet management, real-time remote assistance, and robust maintenance operations; any bottleneck can reduce uptime and increase per‑mile cost.findarticles+1
Strategy and competitiveness	Positions firms as core infrastructure for autonomous mobility, integrating partners like Waymo, WeRide, and others to broaden supply and defend market share.ainvest+3	Intense competition from dedicated AV players (Waymo, others) and from partners themselves; platforms risk being disintermediated or squeezed on margin.findarticles+2
Regulatory environment	Early movers that secure approvals and deploy at scale can lock in data, routes, and brand advantage in key cities.ainvest+2	Regulatory setbacks, accidents, or public pushback can halt deployments, strand capital, and create reputational damage.ainvest+2

The point is that it remains unclear how successful ridesharing companies might be at coping with a shift to robo-taxi alternatives. The sheer difference between an asset-light and asset-heavy business model illustrates the challenges. 

It's Actually Too Early to See Widespread AI Productivity Gains

“Today, you don’t see AI in the employment data, productivity data or inflation data,” says Torsten Slok, Apollo chief economist. “Similarly, for the S&P 493, there are no signs of AI in profit margins or earnings expectations.”

That is not without precedent, as that lag in quantifiable productivity impact also happened when computing technology was applied at work. In fact, it often happens that productivity actually decelerates when a new computing or other general-purpose technology is introduced.

GPTs are "consequential" innovations that transform entire economies over time.


source: MIT

So that J curve is not unusual.

But it might also be the case that productivity measurements are outdated. “It’s possible that “current measures of productivity do not capture the increases in value added that these technologies promote,” the McKinsey co-authors state. “Many new benefits are incorporated into products or services free of charge, for example, which means productivity statistics do not capture them.”

The best available evidence suggests that mismeasurement might explain up to 10 percent of the overall slowdown in productivity growth, a relevant but comparatively small effect,” they say.

Here’s a look at expected productivity gains from artificial intelligence. The impact might be less than you would expect.

source: Apollo Academy

Looking only at generative AI, there are clear and significant time savings, for example.



source: Visual Capitalist



source: Visual Capitalist

But those gains do not translate linearly into firm productivity statistics. Among the reasons: the need to recraft whole business processes (requiring new skills, organizational structure changes).

“General purpose technologies (GPTs) such as AI enable and require significant complementary

investments, including co-invention of new processes, products, business models and human

Capital,” say the authors of a paper published by the National Bureau of Economic Research. “These complementary investments are often intangible and poorly measured in the national accounts, even when they create valuable assets for the firm.”

Also, keep in mind that “whole economy” productivity tends to improve at rates between one percent and two percent annually, over time.

source: St. Louis Federal Reserve

So some economists note that measurable or quantifiable gains from other earlier GPTs took decades, though the impact of computing technologies happened much faster.

source: JP Morgan.

As noted above, AI impact on tasks can be quite high, but the impact on gross national product or productivity will not track in linear fashion. Greater output with similar or less input leads to measurable productivity gains only when the output affects sales and other revenue-related activity.

Isolating the impact of particular inputs requires us to make judgments. When multiple processes change, how do we evaluate the individual impact? If sales channels, production processes, marketing, advertising, applied AI, headcount and customer demand all change at once, any estimation of input factor contribution is subjective.

But in any case, it actually is too early to document AI-driven productivity increases. And the actual impact could well be negative.