a16z: AI is not the end of jobs but the beginning of an era of intelligent inclusivity

Author: David George, a16z General Partner; Source: a16z; Compiled by Shaw; Golden Finance

The anxiety pushed by AI alarmists about a “permanent underclass” of unemployment is fundamentally baseless—and it’s also an old, tired cliché that has resurfaced again with a fresh coat of paint. It’s nothing more than the labor supply fallacy repackaged once more.

The core idea of the labor supply fallacy is: the total amount of work society needs to complete is fixed and unchanging. It assumes today’s existing workers—and other practitioners, machines, and even today’s AI—are locked in a zero-sum game. Under this logic: since the total amount of useful work society can do is constant, the more AI does, the less humans can do.

But this premise is completely at odds with everything we know about human nature, markets, and economics. Human desires and needs have never been fixed. Nearly a hundred years ago, Keynes famously predicted that automation would mean humans only needed to work 15 hours per week. Clearly, he was wrong. He correctly foresaw that automation would create labor surplus, but humans didn’t simply lie down to enjoy leisure; instead, they kept digging into entirely new productive endeavors and enriching their time.

Admittedly, AI will certainly replace some job tasks and compress some occupational roles (there are already signs that this process has begun). Each time a disruptive technology emerges, the labor market landscape is reshaped accordingly—this has always been the case. But to claim that AI will cause permanent, large-scale unemployment across society is nothing more than eye-catching marketing language, faulty economic logic, and a fallacy that ignores historical patterns. On the contrary, productivity improvements actually boost labor demand, because labor itself becomes more valuable.

Below is our complete chain of reasoning.

Humans are already doomed? That’s overstated

We agree with the doomsayers—in fact, anyone with a clear mind can see it: the cost of cognitive labor is falling off a cliff. Tasks that were once considered exclusive to the human brain are increasingly being handled better by AI.

The doomsayers’ argument goes like this: “If artificial intelligence can think for us, then humans’ moat disappears instantly, and humans’ ultimate value becomes zero.” Humans are then entirely replaced. In their view, the thinking work humans need and want has already reached its end; now AI will take on an ever-growing share of mental workload, and humans will gradually become redundant and be phased out by the times.

But the opposite is true: both historical precedent and common-sense logic show that when the cost of a powerful production input drops sharply, the economy never simply grinds to a halt. As costs fall, quality improves, and efficiency accelerates, brand-new products become feasible on the ground, and overall demand expands outward. The Jevons paradox fully applies here.

When fossil fuels made energy cheap and abundant, humans didn’t just eliminate traditional occupations like whalers and lumbermen; we also invented entirely new things like plastics.

Contrary to the doomsayers’ view, we have every reason to believe that AI will produce a similar effect. As AI takes over more and more cognitive work, humans gain freedom to explore unprecedented, bigger-scale, more ambitious frontier fields.

If we look to history for guidance, we can foresee: technological change will ultimately enlarge the entire economic “pie.”

Each economic sector that once held dominance will ultimately give way to newer industries with a larger footprint… and this process will only make the overall economic scale grow even stronger.

Today’s tech sector is already far larger than the old financial, railway, and industrial sectors; yet in terms of share of the overall economy and the market, there is still ample room to grow. Productivity gains are not a negative-sum zero-sum game—they are a positive-sum growth engine that has been heavily empowered. When humans hand large amounts of work over to machines, the result is that the economy and the labor market will only become bigger, more diversified in business models, and more complex in structure.

Doomsayers intentionally ignore the history of human innovation, taking only the present moment when cognitive costs are plunging and treating that instant as the whole ending. They only see AI replacing single job tasks, stop there, and refuse to dig deeper.

“Humans’ mental output will increase tenfold, but we won’t do more thinking or creative work; instead, we’ll just lounge around, take early and long breaks for lunch—everyone will do the same.” This idea is not only wildly lacking in imagination; it is also severely missing even basic real-world observation. The doomsayers package this line of thinking as “realism,” but history has never shown anything like it happening.

The failure of the Luddites

Let’s review history and see what truly happens when step-change productivity breakthroughs hit the economy.

Agriculture

In the early 20th century, before agricultural mechanization became widespread, about one-third of the U.S. labor force worked in agriculture. By 2017, agriculture’s share of employment had fallen to only about 2%.

If automation truly caused permanent unemployment, tractors should have thoroughly destroyed the labor market. But reality was the exact opposite: agricultural output nearly tripled, supporting a large-scale increase in population. Those who left the farms did not suffer permanent unemployment either; instead, they poured into entirely new industries that were previously unimaginable: factories, supermarkets, office buildings, hospitals, laboratories, and later, the service industry and software technology sector.

It’s undeniable that technology did disrupt ordinary farmers’ career development paths; but at the same time, it released an abundance of labor and resources, spawning an entire new economic system.

Electrification

The electrification process follows a similar development logic.

Electrification is not simply a matter of replacing one energy source with another. It replaced traditional drive shafts and belts with independent electric motors, forcing factories to reorganize their operating structures around entirely new production processes, and it also gave rise to entirely new categories of consumer goods and industrial products.

This is the typical characteristic that appears at each stage of technological revolutions. As summarized by Carlota Perez in Technological Revolutions and Financial Capital: in the early stages, huge upfront investments and financial capital chasing after them appear; the cost of durable goods drops significantly; then durable-goods manufacturing enterprises enter a long-term boom lasting across generations.

Electricity also took a long time to release its powerful production-enabling effects. In the early 20th century, only 5% of American factories used electricity to power machines, and even household electrification rates were under 10%.

By 1930, electricity provided nearly 80% of manufacturing power, and over the following decades, the growth rate of labor productivity doubled directly.

Productivity gains not only did not weaken labor demand; they instead drove manufacturing expansion, increased the number of salespeople, expanded credit business, and fostered widespread commercial prosperity. And not to mention the secondary effects of labor-saving equipment such as washing machines and cars: these enabled more people to take on higher-value work that previously had been out of reach.

As car prices fell, both car production and employment exploded.

This is the true role of general-purpose technology: reconstructing economic structures and continuously expanding the boundary of useful work.

Scenes like this repeat throughout history. Did spreadsheet software like VisiCalc and Excel lead to the extinction of bookkeepers? Absolutely not. Technologies that greatly improve calculation efficiency instead caused bookkeeping professionals to surge in number, and they even created an entirely new industry: financial planning and analysis (FP&A).

We reduced about 1 million “bookkeeper” positions, but added roughly 1.5 million “financial analyst” positions.

Brand-new service-sector jobs

Of course, task replacement by technology doesn’t always bring employment growth in neighboring areas of the economy. Sometimes, productivity surplus creates incremental jobs in completely unrelated industries.

But someone might ask: what if AI only makes a small group of people extremely rich, leaving everyone else far behind—then what?

At minimum, these super-rich must spend their wealth, which then invents an entire new service industry out of thin air—this is how history always works:

A big leap in productivity and the wealth it generates creates a large number of brand-new career tracks. Even if, as far back as the 1990s, these careers were technically feasible, without rising household incomes and abundant labor supply, they could never have taken shape.

No matter how people view the various services for the wealthy, the final outcome is that everyone’s lives get better. Because demand expansion pushes the median salary upward dramatically, which in turn creates more people who move into the affluent class.

Stripe economist Ernie Tedeschi gave a highly representative complete case that illustrates how a profession is disrupted, reshaped, and reborn by technology: travel agents.

Did technology reduce demand for travel agencies? The answer is yes—undoubtedly:

Today, the total payroll for travel agencies is only about half of what it was at the beginning of the century, and this is almost entirely caused by technology development.

So does that mean technology kills jobs? Again, the answer is no. Travel agents did not end up in permanent unemployment. They found new work in other parts of the economic system; after excluding the factor of an aging population, the share of the employed population is roughly the same as around 2000.

Meanwhile, those who stayed in the tourism industry empowered by technology benefited from productivity improvements, and their wage levels are actually higher than before.

“During the travel agency industry’s peak in 2000, the average weekly salary of travel agency employees was only 87% of the overall average weekly salary in society. By 2025, this ratio has risen to 99%, meaning that during this period, the wage growth rate of travel agencies outpaced other fields in the private sector.”

So even in this case—although technology indeed disrupted the number of travel agency roles—overall employment among the eligible working-age population remains roughly level compared with before. And the travel agents who remain have seen their income reach historical highs.

Empowerment over replacement (and those yet-to-be-born new jobs)

The final point is crucial, and it once again shows that AI doomsayers only see a small slice of the iceberg.

For some jobs, AI is a disruptive survival threat. But for many more jobs, AI is a capability multiplier that will actually significantly increase the value of those roles. Behind every job facing the risk of being replaced by AI, there’s another set of jobs that will benefit from it.

The AI replacement effect estimated by Goldman Sachs has long been fully offset—and even surpassed—by the productivity gains enabled by AI.

It’s also worth noting that corporate management now clearly values AI enablement over job replacement.

As of now, in corporate earnings call transcripts, the frequency of mentions of “AI enabling efficiency” versus “AI job replacement” is roughly in an 8:1 ratio.

Even though Goldman Sachs doesn’t include software engineers in its “AI enablement roles” list, they may well be the most typical example of an AI efficiency-oriented occupation.

AI is a capability multiplier for programming work. Code submission volume is surging (the number of new applications and new startups is also skyrocketing), while demand for software engineers has turned upward again and is back on a growth path.

Since the beginning of 2025, jobs related to software development have been maintaining growth—both in absolute numbers and as a share of the overall employment market.

Is this the result of AI? To be honest, it’s too early to draw a definitive conclusion. But there is no doubt that AI has greatly empowered software engineering work. What’s more, today every company’s executives list AI as a top priority.

Across industries, businesses are doing everything they can to integrate AI into their own operations. Naturally, this leads to large-scale hiring and deployment to roll out the transformation. This can only increase the value of specialized talent, not reduce it.

AI-related roles are driving wage growth rates that are running ahead of the industry average, especially in the system design field.

At present, the scope of this wage dividend may still be relatively limited, but it is at an extremely early stage. As more professionals in relevant fields continue to increase, employment opportunities will expand as well. In any case, these are not the data points that AI doomsayers want you to see.

Meanwhile, in a well-known tech community, the founder of the newsletter Lenny’s Newsletter, Lenny Rachitsky, points out that the number of hiring roles for product managers (PMs), which had continued to rebound since the industry’s interest-rate environment collapsed earlier, has reached a new high since 2022.

The simultaneous growth in hiring for software engineers and product managers is a great example proving that the labor supply fallacy does not hold. If AI were to replace human cognitive work one-to-one, we would expect to see something like this: “Fewer engineers needed since product managers no longer need as many engineers,” or “Fewer product managers needed since engineers no longer need as many product managers.” But reality is not like that. What we see instead is that hiring demand for both roles continues to warm up. The core reason is that people use AI to accomplish more—much more—work at larger scale.

This is also the fundamental problem with AI doomsayers: lack of imagination. They focus only on scattered job tasks that automation can replace, while ignoring the emergence of brand-new demand frontiers that will spawn entirely new professions that we can’t even imagine today.

Most of the job roles newly created since 1940 did not exist in 1940. In 2000, it was easy to predict that a large number of travel agent jobs would be lost; but it was hard to imagine that, in the future, an entire technology services industry aimed at mid-market enterprises would be spawned around cloud migration—after all, at that time the cloud computing era was still a long way off, and it only truly arrived more than a decade later.

What does the real situation tell us right now?

So far, our analysis has mainly drawn from theory and historical precedent. And regardless of whether it’s theory or past experience, both are favorable to the optimistic camp.

That’s exactly what’s happening. Every time a breakthrough in productivity unlocks potential, it brings demand expansion—or reallocates productivity surplus to other parts of the economy. This means jobs increase: the value of many existing roles rises significantly, and many entirely new careers emerge that we don’t even know about today.

If this time the historical pattern were somehow to break as an exception, then the doomsayers would need to produce rigorous, solid proof—not just panic-mongering and hollow lamentation.

Job displacement does not destroy human civilization; quite the opposite—this logic holds up. Human nature has never been content with the status quo or willing to stagnate. Once we complete something, we pursue the next thing.

Putting aside theory and historical precedent, what does actual data say about the relationship between AI and employment? It should be noted that we are still in an early stage, so the pros and cons have not fully surfaced yet. But existing mainstream data does not support the doomsayers’ narrative. Even from a conservative perspective, data shows there hasn’t been any obvious fluctuation in employment. And more and more emerging data points in the opposite direction: the number of jobs created by AI is far greater than the number it replaces.

First, let’s look at some academic research—not exhaustive, just selecting representative papers from recent years:

• Artificial Intelligence, Productivity, and Labor: Evidence from Business Executives (NBER Working Paper 34984): By synthesizing findings across studies, it shows that although the adoption of artificial intelligence has not yet caused a significant change in the overall size of employment, it has already begun reshaping task allocation within firms and the structure of occupations. Specifically, routine clerical and administrative work is more easily replaced by AI; while analytical, technical, and managerial work tends to be complemented and empowered by AI.

• Firm-level artificial intelligence-related data (Federal Reserve Bank of Atlanta Working Paper 2026-3): Based on four survey data sets, on average, more than 90% of firms believe that over the past three years, artificial intelligence has had no impact on their employment.

• 

• The microstructure of AI diffusion: Evidence from firms, business functions, and employee work tasks (U.S. Census Bureau, Center for Economic Studies, CES 26-25 Working Paper): The study finds that, in aggregate, employment changes triggered by AI applications remain relatively mild. Only about 5% of AI-applying firms report that employee headcount has been affected. Among these, the share of firms reporting job increases (firm weight 2.3%, employment headcount weight 3.7%) and job decreases (firm weight 2.0%, employment headcount weight 2.4%) are almost evenly split.

• 

• Tracking AI’s impact on the labor market (Yale Budget Lab, April 16, 2026): “Although there is widespread societal concern that AI will disrupt the labor market, our data indicates that these concerns are largely subjective guesses. Based on the results shown by the existing data, AI’s overall impact on the labor market is mainly stable and has not caused a disruptive shock at the macroeconomic level.”

The logic is easy to understand. Recent research repeatedly conveys the same conclusion: overall employment shows no net change, but there is a structural reconfiguration between jobs and work tasks. Some studies also show that the overall net effect of applying artificial intelligence to hiring is positive.

Beyond the conclusion that “there is no obvious overall change,” there is one noteworthy exception. Studies from Stanford University, the Federal Reserve Bank of Dallas, and the U.S. Census Bureau all find (to varying degrees) that entry-level jobs with high AI penetration are becoming increasingly difficult to find.

However, before drawing the hasty conclusion that “AI is destroying entry-level jobs,” it’s important to point out that these studies also broadly find that the number of entry-level jobs empowered by AI is increasing—and entry-level jobs that are completely unaffected by AI are also increasing.

Even if we temporarily assume that certain entry-level jobs are indeed being replaced by AI—rather than being caused by macro hiring-cycle fluctuations, workers aging and staying longer in roles, and other factors—looking at the bigger picture, the data is already fairly clear in showing that: the overall effect of AI on employment is approaching zero.

This may be the most concise summary of the current state of AI’s impact on employment.

“Currently, no statistically significant correlation has been found between artificial intelligence and the unemployment rate or the employment growth rate.”

There may be two coexisting trends at the market level: on one hand, talent flows into AI-empowered roles; on the other, AI-replacement roles are pushing out incumbent workers.

In industries where AI enablement is stronger, hiring growth is faster and unemployment is lower; while in industries with higher AI replacement risk, it’s the opposite.

In other words, employment overall remains neutral, but the internal structure has already changed: some jobs are disappearing while others are emerging; some roles become less valuable, while other roles see their value rise and the premium becomes visible. Based on the current growth rate, demand for programmers will surpass pre-pandemic levels in less than two years. AI even, on its own, has stabilized the San Francisco office market.

This loops back to our original view: AI is destined to eliminate and compress some jobs and business models, but if you think that’s the whole story, you’re dead wrong. When facing a transformative technology, what we should expect to see is the labor market re-adjusting and ultimately growing—not collapsing into mass unemployment. This pattern has been proven repeatedly by history, and this time is almost certain to repeat as well, with the trend already underway.

Knowledge work is just beginning

This is a cliché, but the point is still valid: this is not the end of knowledge work; it’s just the beginning.

Automation strips away repetitive, low-level work and pushes human labor toward higher-value levels. The logic behind it is simple: humans are naturally driven to expand outward. When a scarcity at one layer is broken, people move to new demands at a higher layer. When food costs drop, people shift more spending toward housing, healthcare, education, travel, entertainment, life convenience, pets, safety, beauty and anti-aging, and other areas.

The labor market works the same way. New careers will keep emerging because people’s ambition knows no bounds; conquering old domains creates entirely new frontiers waiting to be explored.

Currently, the number of new business registrations has exploded, and it shows a strong positive correlation with AI deployed applications.

The number of new app releases on application stores has a year-over-year growth rate as high as 60%.

There is no reason to view the modern economy as a museum of existing jobs and old career structures. On the contrary, it is a creative resource-allocation machine that continuously creates new jobs, new work, new goals, and new innovations.

For a long time, robotics technology was mostly treated as science fiction, because the computational requirements in dynamic environments were too high. But now, AI is turning an entirely new robot industry from concept into reality—moving it into the real world and the range of what we can actually see.

The scale of robotics datasets is growing explosively. In just two years, the industry’s share jumped from tenth place to first.

A huge number of robot-related roles previously had no market demand—until AI unlocked this latent demand.

Let’s emphasize again: none of the above means that all existing jobs can survive intact. The U.S. Bureau of Labor Statistics expects that roles such as customer service representatives and medical transcriptionists will shrink—and this shrinking trend may have already started.

Some jobs will disappear, while other job categories will shrink in scale. The industry landscape must adjust, and the transition will come with pain. Productivity gains penetrating into the entire economic system often takes time as well. We should empathize with the disruptions brought by such changes and do everything possible to push the transition toward a smooth and steady passage, including proactively carrying out job retraining—something a16z is also very willing to support.

The original purpose of productivity progress is to help humans escape dull and heavy mechanical labor, and this AI revolution will be no exception.

But so-called “AI job apocalypse” only holds under one ridiculous assumption: that human desires and innovative thinking will completely come to a halt the moment intelligent technologies become cheap and widely adopted. Obviously, that’s nonsense. Personally, I do not subscribe to a “Wall-E”-style doomsday narrative of lying down and giving up. And I believe there are many others who share this view too.

From a macro perspective, the future is absolutely not one of everyone being unemployed, everyone settling into comfort, and wasting life away relying on streaming entertainment and motorized rides for transportation.

The future looks like this: the cost of intelligence drops dramatically, the market size keeps expanding, new companies keep emerging, and new industries are born in sequence. Humans will engage in higher-level creative work. The total amount of work has never been fixed, and the demand for cognitive labor has never been fixed either—this was never true in the past, and it will never be true in the future. AI is not the endpoint of labor; it is the beginning of the era of intelligent inclusion.