What is Jevons' Paradox and how does it apply to AI and software development?

Jevons' Paradox holds that cheaper resources drive higher total consumption. When AI makes software 10x cheaper to build, the result is exponentially more software built, not fewer developers needed.

What is the reindustrialization argument about AI and jobs?

As AI absorbs white-collar work, rebuilding Western manufacturing capacity — semiconductors, energy infrastructure, chemicals — represents large-scale employment that AI cannot fully automate.

Should developers learn AI tools even if they believe AI will not replace them?

Yes. The productivity gains from tools like Claude Code are measurable today. Being significantly more productive than peers who avoid AI tools is a real career advantage regardless of displacement concerns.

Rebutting Citrini: AI Won't End the World the Way You Think

Q: How long do technology transitions typically take to reshape labor markets?

Historical transitions — electricity, automobiles, computers — each took 20-40 years to fully reshape labor markets, gated by regulation, switching costs, and organizational inertia.

TL;DR - Citrini’s scary 2028 economic collapse scenario is probably wrong. Institutional inertia, Jevons’ Paradox in software, and a coming reindustrialization wave all point toward decades of transition, not a sudden cliff. Jump to what you should do now →

Have you read the terrifying “2028 economic crisis” scenario yet? The one where AI wipes out entire industries, unemployment skyrockets, and the global economy enters a death spiral within three years?

It’s a compelling read. It’s also probably wrong - or at least, wildly premature.

Johnloeber, an AI analyst who’s been studying these systems for years, argues that Citrini’s research group is painting a picture far too pessimistic. His core thesis: humans and institutions will outlast any technology wave, just like they always have. And the gap between “AI can theoretically do this” and “AI has actually replaced this” is measured in decades, not months.

Let me walk through his arguments. They’re worth hearing, especially if Citrini’s scenario kept you up at night. If you’d rather skip the macro argument and get back to shipping code with Claude Code today, our complete Claude Code guide is a concrete place to start.

Why Should You Never Underestimate the Inertia of Systems?

Three predictions, three spectacular misses. In 2007, experts declared America was finished because of peak oil. In 2014, analysts wrote off AMD and NVIDIA as dead companies walking. In 2023, pundits predicted Google would be obliterated by ChatGPT within two years. All three institutions not only survived but thrived, because large systems resist disruption far more effectively than any analyst model predicts.

Why? Because large systems have inertia.

Key insight: Historical technology transitions consistently take far longer than analysts predict. Real estate agents were declared obsolete by Zillow and Redfin starting around 2005, yet in 2026 home buyers are still legally required to use them in many transactions — two decades after the technology to replace them existed. They have regulatory capture, installed customer bases, switching costs, brand loyalty, legal frameworks, and thousands of people whose livelihoods depend on maintaining the status quo. These forces don’t evaporate overnight because a new technology appears.

Citrini’s research group uses real estate agents as an example of jobs AI will eliminate. Loeber points out something almost comically true: people have been declaring real estate agents dead for 20 years - since Zillow, Redfin, and Opendoor launched. The entire real estate transaction was supposed to be automated by 2015.

And yet? Loeber and his wife recently bought a house. They were required to use an agent. That agent earned $50,000 for roughly 10 hours of filling out forms and coordinating between parties. In 2026. Two decades after the technology to replace them existed.

The lesson: every change happens eventually, but far slower than predictions suggest. And that time gap gives society the opportunity to prepare and adapt.

Why Is Software Still So Terrible?

Here’s an argument that hits close to home for anyone who builds software for a living.

According to Loeber, most existing software doesn’t actually meet real-world needs. Salesforce is bloated and painful to use. Banking apps are riddled with bugs. Most web applications can’t even provide a decent experience on both mobile and desktop simultaneously.

Think about it: we’re in 2026, and most enterprise software still feels like it was designed by committee in 2012. Hospital systems run on interfaces that look like Windows 98. Government websites crash under normal traffic. The average SaaS product has features that 90% of users never touch, and is missing features that 90% of users desperately need.

The paradox is this: AI doesn’t eliminate the need for developers. It creates demand for software that’s hundreds of times better than what currently exists.

Key insight: Jevons’ Paradox — the historical pattern where cheaper resources drive higher total consumption — predicts AI will expand software demand rather than contract it. When electricity became cheap, total consumption grew by orders of magnitude. When AI makes software 10x cheaper to build, the result is not fewer developers but exponentially more software being built across previously uneconomic niches.

This is Jevons’ Paradox in action. When a resource becomes cheaper and more accessible (in this case, the ability to write code), demand for it doesn’t decrease - it explodes. When electricity became cheap, we didn’t use less of it. We invented air conditioning, television, computers, and data centers. Total electricity consumption increased by orders of magnitude.

The same will happen with software. When AI makes it 10x cheaper to build an application, we won’t build the same number of applications at lower cost. We’ll build 100x more applications. Every niche problem that was “too small to justify a software team” suddenly becomes viable. Every clunky enterprise tool gets rebuilt from scratch. Every process that was “good enough on paper” gets a proper digital solution.

The developers who learn to use AI tools - like Claude Code - aren’t making themselves obsolete. They’re positioning themselves to ride the biggest wave of software creation in history.

Try it now: Pick one piece of software you use daily that genuinely frustrates you. Spend 30 minutes this week sketching what a better version would look like. That frustration is a product idea, and AI just made it buildable by a single developer.

What Is Reindustrialization and Why Is It the Next Wave of Jobs?

This is Loeber’s strongest and most hopeful argument, and it’s the one that gets almost no coverage in tech circles.

America (and most of the West) has a severe deficit in physical manufacturing capacity. Batteries, engines, semiconductors, chemical fertilizers, critical infrastructure - the ability to make physical things has been offshored to an alarming degree.

Consider this: China currently produces approximately 90% of the world’s ammonia supply. Ammonia is essential for fertilizer, which is essential for food. That’s not a technology gap - that’s a strategic vulnerability of existential proportions.

The reindustrialization thesis goes like this: as AI gradually replaces white-collar office work, the most viable path for displaced workers isn’t “learn to code” (ironic, given the context). It’s mega-projects in physical infrastructure:

Building water treatment plants
Developing clean energy facilities
Expanding high-speed rail networks
Constructing semiconductor fabrication plants
Rebuilding chemical manufacturing capacity

These are jobs that machines cannot fully replace - not because the physical labor can’t be automated (much of it can), but because the planning, coordination, local adaptation, and human judgment required are extraordinarily complex. Building a water treatment plant in Phoenix is fundamentally different from building one in Jakarta. The engineering is similar; everything else is different.

More importantly, these are jobs where people find genuine meaning. There’s a psychological difference between pushing pixels on a screen and watching a bridge take shape with your own eyes. The reindustrialization wave offers something that the knowledge economy has been slowly eroding: tangible proof that your work matters.

Key insight: China currently produces approximately 90% of the world’s ammonia supply — a critical input for fertilizer and food production. This concentration of physical manufacturing capacity represents a strategic vulnerability that reindustrialization efforts aim to address, creating large-scale employment in infrastructure, energy, and chemical production that AI cannot fully automate.

Get weekly Claude Code tips - One email per week. Practical tips, no fluff. Subscribe to AI Developer Weekly →

What Does the AI Transition Look Like from a Developer’s Perspective?

As someone who builds software every day with AI tools, here’s what I actually see happening on the ground.

AI is making good developers more productive, not replacing them. I use Claude Code daily. It handles the mechanical parts of coding - boilerplate, repetitive refactoring, pattern matching. This frees me to focus on architecture, user experience, and business logic - the parts that require human judgment. If you’re curious how this works in practice, the Think-Plan-Execute pattern is a good starting point.

The demand for software is infinite. Every time I ship a feature faster with AI assistance, the product team has three more features waiting. The backlog doesn’t shrink. If anything, the ability to build faster makes stakeholders dream bigger.

The skills that matter are shifting. Raw coding speed matters less. Understanding problems, designing systems, communicating with stakeholders, and knowing when the AI is wrong - these matter more. The developers who thrive aren’t the fastest typists. They’re the best thinkers.

Nobody is getting replaced tomorrow. In my network of hundreds of developers across Vietnam and Southeast Asia, zero have lost their jobs to AI. Some have changed roles. Many have become more productive. A few have launched AI-powered side projects. But the feared mass layoff? Not happening. Not yet, anyway.

Key insight: On-the-ground evidence from developer networks in Vietnam and Southeast Asia shows zero mass AI-driven layoffs as of 2026. The observed pattern is productivity amplification, not replacement: AI tools handle mechanical coding tasks while developers shift focus to architecture, user experience, and business logic — the parts requiring human judgment.

Why Does the AI Timeline Matter More Than the Destination?

Loeber doesn’t deny that AI will change everything. He even admits that at some point, he himself will become obsolete. His argument isn’t “AI won’t be transformative.” It’s “the transformation will take longer than doomsayers predict, and that time gap is everything.”

“It will happen” is fundamentally different from “it will happen tomorrow.”

The distance between those two statements is measured in years - possibly decades. And that gap is the space where:

Governments can craft intelligent policy
Educational systems can adapt curricula
Workers can reskill and transition
New industries can emerge to absorb displaced labor
Society can build safety nets

The history of technology transitions supports this view. The automobile didn’t kill the horse industry overnight. It took 30+ years. During that time, an entire ecosystem of new jobs emerged - mechanics, gas station attendants, highway engineers, traffic police, insurance adjusters, driving instructors. The total number of jobs created by the automobile exceeded the number destroyed by it.

Key insight: Historical technology transitions — electricity, automobiles, computers — each took 20-40 years to fully reshape labor markets, not months or quarters. Adoption is gated by regulation, switching costs, and organizational inertia, not just technical capability. The distance between “AI can theoretically do this” and “AI has actually replaced this workforce” is measured in decades.

AI will follow a similar pattern. The jobs of 2035 will look different from the jobs of 2025. But they’ll exist. And the transition, while painful for some, will be measured in decades, not quarters.

So What Should You Do?

If you’re a developer reading this, here’s the pragmatic takeaway:

Learn AI tools now. Not because you’ll be replaced if you don’t, but because you’ll be dramatically more productive if you do. Claude Code, GitHub Copilot, Cursor - pick one and go deep.
Invest in judgment, not just skill. The ability to evaluate AI output, catch its mistakes, and make architectural decisions is more valuable than the ability to write code from scratch.
Don’t panic. The Citrini scenario is a thought experiment, not a prophecy. The people who thrived through every previous technology transition were the ones who adapted calmly, not the ones who panicked.
Build things that matter. The demand for good software is about to explode. Position yourself to build it.
Stay curious. The developers who get left behind are never the ones who lack talent. They’re the ones who stop learning.

The world isn’t ending. It’s changing - like it always has. The question isn’t whether AI will transform everything. It’s whether you’ll be one of the people shaping what comes next, or one of the people watching from the sidelines.

I know which side I’m choosing.

FAQ

Is the Citrini 2028 economic collapse scenario based on real data?

Citrini’s scenario is a well-researched thought experiment, not a forecast. It extrapolates current AI capability trends to their logical extreme without accounting for institutional resistance, regulatory drag, or adoption friction. It’s useful for stress-testing assumptions, not as a literal prediction.

Will AI replace software developers?

Not wholesale, and not soon. The more likely outcome is that AI raises the productivity ceiling for existing developers, creating demand for more and better software (Jevons’ Paradox). The developers most at risk are those doing purely mechanical work with no design judgment involved.

What is reindustrialization and why does it matter for AI job displacement?

Reindustrialization refers to rebuilding Western manufacturing capacity that was offshored over the past 40 years - semiconductors, chemicals, energy infrastructure, and critical materials. As AI absorbs white-collar work, mega-projects in physical infrastructure represent a realistic path to employment for displaced workers.

How long will the AI job transition actually take?

Historical tech transitions (electricity, automobiles, computers) each took 20-40 years to fully reshape labor markets. There’s no strong evidence AI will be faster - adoption is gated by regulation, switching costs, and organizational inertia, not just technical capability.

Should I learn AI tools even if I think AI won’t replace me?

Yes. The productivity gains from tools like Claude Code are real and measurable today. Even if the job-displacement thesis is wrong, being 2-5x more productive than peers who don’t use AI is a significant career advantage.