a16z合伙人驳斥AI末日论：别慌，技术变革会做大蛋糕

Original author: David George

Original translation: Felix, PANews

Editor's Note: The current AI "doomsday" narrative seems to dominate public opinion, with fears of "AI taking jobs" and "unemployment" spreading globally. People from all walks of life are also offering suggestions for the disruptive changes AI is about to bring. However, a16z General Partner David George argues that the "doomsday" view is completely unfounded, lacking evidence and imagination, and failing to understand humanity. The following is the full text.

The "permanent underclass" rhetoric proposed by AI alarmists is unconvincing. This is nothing new; it's just the "lump of labor fallacy" in a new package.

The "lump of labor fallacy" claims that the total amount of work needed in the world is fixed. It assumes a zero-sum game between existing workers and anyone or anything that could do the same work, be it other workers, machines, or now AI. If the total amount of useful work needed is fixed, then if AI does more, humans must necessarily do less.

The problem with this premise is that it contradicts everything we know about people, markets, and economics. Human needs and desires are by no means fixed. Nearly a century ago, Keynes predicted automation would lead to a 15-hour work week, but history proved him wrong. He was correct about automation causing a "surplus of labor," but instead of resting on our laurels, we found new and different productive activities to fill our time.

Of course, AI will absolutely eliminate some jobs and compress certain roles (and there is evidence this may already be happening). The landscape of the labor market will change, as it inevitably does with every transformative technology. However, the idea that AI will lead to economy-wide, permanent unemployment is bad marketing hype, poor economics, and historical ignorance. On the contrary, increased productivity should increase the demand for labor, as labor becomes more valuable.

Here are our reasons.

"Humanity is Doomed?" No Joke

We agree with the "doomsday" crowd on one point: the cost of cognition is plummeting. AI is becoming increasingly proficient at tasks that were, until recently, considered the exclusive domain of the human brain.

The "doomsday" crowd argues: "If AI can think for us, humanity's 'moat' disappears, and our ultimate value goes to zero." Humanity is finished. Apparently, we have already done all the thinking that needs or wants to be done, and now AI will take on an increasing cognitive load, leading humans towards obsolescence.

However, the truth is: precedent (and intuition) shows that when the cost of a powerful input drops, the economy does not stagnate. Costs fall, quality improves, speed increases, new products become viable, and demand expands outward. Jevons paradox applies once again. When fossil fuels first made energy cheap and abundant, we didn't just make whalers and woodcutters unemployed; we invented plastic.

Contrary to the "doomsday" view, we have every reason to expect a similar impact from AI. Since AI will take on an increasing cognitive load, humans can free up their hands to explore new, grander frontiers than ever before.

History shows that technological change will inevitably make the economic pie larger.

Each "dominant economic sector" was replaced by a larger successor... which in turn further expanded the economy.

Today's tech sector is far larger than finance, railroads, or industry, yet it still represents a small portion of the overall economy or market. Productivity gains are far from a negative-sum game; they are a powerful positive-sum force. Delegating so much work to machines ultimately leads to a larger, more diverse, and more complex economy and labor market.

The "doomsday" crowd wants you to ignore the history of innovation, focus only on the sharp decline in cognitive costs, and take that as the whole truth. They see task substitution and stop thinking.

"We will increase cognitive output tenfold, but instead of doing more thinking, we'll pat our bellies and head off for an early lunch, and everyone else will do the same." This statement reflects not only a severe lack of imagination but also a failure to observe basic facts. The "doomsday" crowd calls it "realism," but it is simply impossible.

The Failure of Luddism

(PANews Note: Luddism refers to a social movement in early 19th century England where workers protested against the Industrial Revolution and destroyed industrial machinery to protest worsening working conditions and unemployment.)

Let's look at what actually happens when a massive productivity leap sweeps through the economy.

Agriculture

At the beginning of the 20th century, before the widespread adoption of agricultural mechanization, about one-third of the employed population in the United States worked in agriculture. By 2017, this figure had dropped to about 2%.

If automation causes permanent unemployment, the tractor should have completely destroyed the labor market. But it didn't. Agricultural output nearly tripled, supporting significant population growth, and instead of becoming permanently unemployed, these workers flooded into previously unimaginable industries, factories, shops, offices, hospitals, laboratories, and eventually into the service and software sectors.

So, while it's true that technology disrupted the career prospects of the average farm worker, it also unleashed a surplus of global labor (and resources), giving birth to a completely new economic system.

Electrification

The story of electricity is similar.

Electrification wasn't just swapping one energy source for another. It replaced drive shafts and belts with independent electric motors, forcing factories to reorganize around entirely new workflows and creating entirely new categories of consumer and industrial goods.

This is exactly what we expect at different stages of technological revolution, as Carlota Perez documented in "Technological Revolutions and Financial Capital": massive upfront investment and financial interests, a sharp drop in the cost of durable goods, and a subsequent generational boom for durable goods manufacturers.

Electricity didn't realize its productivity advantages overnight. In the early 20th century, only 5% of U.S. factories used electric power to drive machinery, and less than 10% of homes were electrified.

By 1930, electricity supplied nearly 80% of manufacturing power, and labor productivity doubled over the following decades.

Far from diminishing the demand for labor, increased productivity brought more manufacturing, more salespeople, more credit, and more business activity, not to mention the ripple effects from labor-saving devices like washing machines and cars. These devices allowed more people to engage in higher-value work that was previously unattainable.

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

This is what true general-purpose technologies do: they restructure the economy and expand the frontier of useful work.

We see this time and again. Did VisiCalc and Excel end the career of bookkeepers? Absolutely not. The dramatically more efficient computing technology instead led to a surge in the number of bookkeepers and spawned the entire Financial Planning & Analysis (FP&A) industry.

We lost about 1 million "bookkeepers," but gained about 1.5 million "financial analysts."

New Service Sector Jobs

Of course, job displacement doesn't always lead to job growth in related economic fields. Sometimes, productivity gains translate into new jobs in entirely unrelated industries.

But what if AI means some become incredibly wealthy, while others are left far behind?

At the very least, those ultra-wealthy individuals will have to spend their money somewhere, just as they did before, creating entirely new service industries from scratch:

Massive productivity gains and the ensuing wealth creation spawned entirely new fields of work that might never have emerged without income growth and increased labor supply (even though these fields were technically feasible long before the 90s). Regardless of one's views on service industries catering to the wealthy, the ultimate outcome benefits everyone because increased demand leads to significant rises in median wages (thus creating more "wealthy" people).

Stripe's in-house economist Ernie Tedeschi provides a comprehensive case study of how technology disrupted, transformed, and reshaped the travel agent profession.

Did technology reduce the demand for travel agents? The answer is yes.

Today, the number of travel agent employees is roughly half of what it was around 2000, almost certainly due to technological advancements.

So, does this mean technology killed jobs? No, because travel agents didn't become permanently unemployed. They found work in other sectors of the economy, and today's overall employment-to-population ratio for the working-age population is roughly the same as in 2000 (adjusted for an aging population).

Meanwhile, for those who remained in the now technology-empowered travel agent industry, increased productivity means higher wages than before:

"At their peak in 2000, travel agents' average weekly wages were 87% of the overall average weekly wage. By 2025, this figure had reached 99%, meaning travel agents' wages grew faster than other private sectors during this period."

Therefore, even if technology did impact travel agent employment, the overall employment rate for the working-age population remained stable, and the remaining travel agents are better off than ever.

Augmentation > Substitution (and Jobs Yet to Appear)

This last point is very important and shows again that the "doomsday" crowd only tells a small part of the story.

For some jobs, AI poses an existential threat. This is true. But for other jobs, AI acts as a multiplier: making those jobs more valuable. For every job at risk of AI substitution, there are other jobs poised to benefit:

Goldman Sachs estimates that the "AI substitution" effect is far smaller than the "AI augmentation" effect.

It's also worth noting that management teams seem to focus more on augmentation than substitution:

To date, mentions of "AI as an augmentation function" on earnings calls outnumber mentions of "AI as a substitution function" by about 8 times.

Although Goldman Sachs doesn't even include software engineers on its "augmented" list, they are perhaps the best example of AI-augmented talent.

AI is a multiplier for coding. Not only have git pushes surged (along with the creation of new apps and businesses), but the demand for software engineers also seems to be rising:

Since early 2025, software development job openings (both in number and as a percentage of the overall job market) have been growing steadily.

Is this related to AI? Frankly, it might be too early to tell, but AI undoubtedly enhances software engineering productivity, not to mention that AI has become a focal point for executives at every company.

Given everyone is trying to figure out how to integrate AI into their own businesses, it's not surprising that companies are hiring aggressively, which undoubtedly increases the value of some employees rather than diminishing it.

The proliferation of AI appears to be driving above-average wage growth (particularly in systems design).

These gains might be limited currently, but it's still early days. As expertise expands, opportunities will increase. Regardless, this is not the data the "doomsday" crowd wants you to see.

Meanwhile, according to Lenny Rachitsky (founder of Lenny's Newsletter, a platform for insiders in the tech world), the number of open project manager positions continues to rise (after a significant decline due to interest rate fluctuations), and is now higher than at any point since 2022:

The growth in hiring for software engineers and product managers is powerful evidence supporting the invalidity of the 'lump of labor fallacy'. If AI completely replaced human thinking ability, you might expect "engineers needing fewer product managers" or vice versa, but that's not what's happening. We see robust and sustained demand for both types of talent, precisely because people are becoming more efficient.

This is why the "doomsday" crowd's claims are fundamentally a failure of imagination. They only focus on the jobs that will be automated away, ignoring the new demand areas that are about to create jobs we haven't even conceived of yet:

Most jobs created since 1940 didn't even exist in 1940. By 2000, it was easy to imagine travel agents becoming obsolete, but much harder to envision a mid-market tech services industry built around "cloud migration," since widespread cloud adoption was still over a decade away.

What Does the Current Situation Look Like?

So far, we've mainly discussed theory and precedent, because both support the optimists:

That's right. Every productivity increase has led to demand growth, or the reallocation of surplus resources to other parts of the economy. This means more jobs, many of which are significantly more valuable, along with jobs never heard of before. If this time is different, the "doomsday" crowd needs to provide stronger arguments than mere empty rhetoric.

It makes perfect sense that "job displacement" is not the end of civilization (quite the opposite, in fact). Human nature is to be restless. We finish one piece of work and look for another.

But, setting aside theory and precedent for a moment, what do the actual data say about AI and employment? While it's still early days (for better or worse), current data does not support the "doomsday" view. If anything, the trend is 'no significant change,' but there is emerging data pointing in the opposite direction: AI is creating more jobs than it is taking away.

First, let's look at some academic research. This is not an exhaustive literature review, just a few examples of recent papers:

• "AI, Productivity, and the Labor Force: Evidence from Business Executives" (NBER Working Paper 34984): "Taken together, these results suggest that while AI adoption has not yet led to significant changes in total employment, it has begun to reshape the allocation of tasks and occupations within firms. In particular, routine clerical and administrative activities appear more susceptible to substitution, while analytical, technical, and managerial tasks