Is Artificial Intelligence Overhyped?

An economist's take on AI hype, bubble, jobs and skills

Is AI Overhyped? An Economist’s Take on What We’re Actually Missing

For my latest episode of I’ve Got Questions, I sat down with leading AI economist and professor Ajay Agrawal to explore whether AI is overhyped, why businesses are seeing low productivity gains, how this technology will transform the workforce, and what the transition from a workforce centered around jobs to a workforce centered around skills will look like.

We’re Overestimating the Speed, Not the Impact

Professor Agrawal reframes the discussion around AI hype entirely: AI isn’t overhyped, but we are overestimating the timing on its transformation and how fast companies will adapt.

His data show that AI transformation will be closer to the Industrial Revolution, which caused GDP growth to shoot up drastically after 1750, than the internet, which has contributed notable but relatively moderate GDP growth. “It seems very hard to imagine that AI will not be transformational to every corner of the economy.”

But the speed of AI transformation could be much slower than many are expecting.

Professor Agrawal draws a direct parallel to electricity. Twenty years after electricity was invented, only 3% of companies had adopted it, and they weren’t seeing major economic gains. Why? Because the early perceived value—slightly reducing operational costs—wasn’t enough to justify tearing apart existing infrastructure.

The massive productivity lifts, sometimes 300-500%, only came later. Companies had to realize they could completely redesign their factory floors, getting rid of heavy support columns and building cheaper, single-story factories. This process is called co-invention: reimagining your entire operation around the new technology.

So right now companies are in the early stage of AI where they’re trying to apply it to existing workflows and business models. This won’t lead to very notable productivity improvements. Once the co-invention starts, where companies redesign business models and infrastructure around AI—which could take years to decades if we’re looking at electricity as a model—we’ll start to see the real economic transformation.

Why Waiting Is Not an Option

Can companies afford to wait and see how AI evolves? According to Professor Agrawal, this would come at a significant cost. AI is the first technology in history that learns from use.

This means companies cannot afford to wait. The AI is constantly improving. If your organization writes off AI as hype, you’re creating a compounding disadvantage. Even if all AI progress stopped tomorrow, the current capabilities are significant enough to cause fundamental disruption across every industry.

The companies that start experimenting now are teaching their AI systems to understand their specific problems, data, and workflows. The companies that wait will be competing against systems that have been learning for years.

But What About a Bubble?

Even if AI proves to be the most transformative technology humanity has ever seen, could a bubble still burst in the short run? Think the dot-com crash or the electricity bubble of the 1880s.

One concerning area for Agrawal is the enormous capital being invested in data centers. Billions of dollars are flowing into infrastructure representing bets on future demand for training and using AI models. Some are asking if data centers are the new railroads—infrastructure that countries invested in too early, leading to financial crashes because actual use lagged behind investment.

Agrawal’s advice? Anticipate “ebbs and flows”—periods of hype and excitement followed by disillusionment—while maintaining the long-term expectation that the technology will change the game.

The Real Story Behind the College Graduate Crisis

I raise the recent concerns around the college graduate crisis. Stanford research showing a 6% decline in employment for young workers in AI-exposed fields like finance, marketing, and computer science. Another study claiming a 22% decline in employment for new college graduates overall.

The fear: AI systems are displacing human workers at entry level, doing the tasks new employees typically handled.

But Agrawal points out crucial context that’s being overlooked. The “Canary in the Coal Mine” paper found that overall employment actually went up in the occupations studied. Even more curious: while the number of young employed people decreased, their wages did not fall. For a labor economist, this is strange—falling demand should cause wages to drop.

One explanation? Young workers are finding employment in new places, like startups, that aren’t captured in the datasets measuring mid- and large-sized companies. The authors themselves titled their paper with a question mark, signaling the data isn’t conclusive evidence of catastrophic shift.

The takeaway for new graduates: if you can immediately start building experience and judgment in your field—even outside traditional pathways—you may surpass those who rely solely on conventional credentials.

What Actually Changes About Work

“The skills that gave you dominance before AI may not be the same skills that give you dominance after AI.”

Take law or journalism, where being able to write well used to be essential. Now that AI can write better than most people, competition moves “upstream.” The real competition is now about how well you can think: scenario planning, anticipating moves, deep analysis, building trust, crafting the story.

The single most important skill for the future, Agrawal believes, is judgment. This is non-negotiable. Judgment involves deciding what information to give the AI, interpreting whether the output is good, and weighing trade-offs. Your job becomes directing AI systems by applying your judgment.

This means skill sets and organizational charts could look radically different. Someone with great judgment in marketing could potentially move to finance, because their ability to assess trade-offs is now more valuable than specific technical capability.

I highly recommend listening to this part of the episode to understand how to develop judgment as a skillset.

Economic Expansion, Not Just Destruction

While many fear widespread job destruction, Professor Agrawal views it differently: when the cost of cognitive tasks drops—legal services, healthcare analysis, financial planning—we will demand more of them.

If legal services become much cheaper, we’ll use them for things we never would have before. If healthcare analysis drops to a fraction of the cost, more people get access and demand doesn’t max out. This is called demand elasticity, and it means AI isn’t just a zero-sum game.

Falling costs lead to economic expansion and the ability to do more things. While the specifics of every job will change, the long-term impact isn’t a world of “less thinking.” Applying judgment requires significant thought.

Finally, we discussed the impact of humanoid robots and whether a post-work world could eventually be in the cards.

I learned so much from this conversation. I would love to hear what you think

.