Post-Labour Economy: GDP Grows While Jobs Vanish

Adapted from my fireside chat at South Summit Brazil 2026 with Christopher Locke

I run three product lines. AI does the work that used to require a team of twenty-plus people. I'm talking about the actual work: drafting sales proposals, analyzing competitive intelligence, managing CRM pipelines, writing code, processing customer onboarding, scheduling, email ops. Not the fluff tasks. The real ones.

My "Chief of Staff" agent preps every meeting by pulling the other person's LinkedIn history, cross-referencing past notes, and building a briefing document. My sprint planning agent converts Slack conversations and meeting transcripts into prioritized task boards. My content system drafts, evaluates, and publishes across three channels without a single human editor.

I don't say this to brag. I say it because this gives me a specific vantage point for the question Christopher asked me on stage in Porto Alegre: What happens when paid employment stops being central to economic life?

I'm not theorizing. I've already run the experiment on my own company. And what I found is more unsettling than the optimists admit, and more tractable than the pessimists claim.

Why the Luddite Fallacy Has an Expiration Date

For 200 years, every time someone warned about machines taking jobs, economists had a reliable response. The Luddite Fallacy. Destroy old jobs, create new ones, net positive. The cotton gin killed hand-ginning jobs but created cotton shipping, textile manufacturing, railroad logistics. The internet killed travel agents but created web developers, social media managers, SEO specialists.

The pattern held because of a single underlying assumption: humans were the only viable option for cognitive work. Machines could out-muscle us. They couldn't out-think us. That constraint guaranteed that as old tasks automated, new tasks requiring judgment, creativity, and communication would appear, and only humans could fill them.

That constraint no longer exists.

Anthropic's CEO projects that AI will be able to do "effectively anything that any human can do from behind a keyboard" within 2-3 years. That's not a fringe prediction. It's from the company building one of the three most capable AI systems on the planet. Google DeepMind gives "roughly even odds on AGI" by the end of this decade. OpenAI targets a fully automated AI researcher by 2028. These timelines are converging across every major lab.

And the capability curve backs it up. According to METR's trajectory analysis, agent task horizons are doubling every 4-7 months. Systems now run 14-hour autonomous coding sessions on SWE-bench style tasks, up from a 2-hour baseline a year ago. Claude and GPT-4 class models score above 90% on GPQA, a graduate-level science benchmark designed to stump PhD holders. Six years of uninterrupted improvement with no sign of plateauing.

I used to hire people for tasks I now solve with Claude in twenty minutes. Content drafts, competitive research, data analysis, code review. These weren't clerical tasks. They required judgment. And AI handles them at a quality level that would have gotten someone promoted three years ago.

Our AI Maturity Index dataset tells the story quantitatively. We've analyzed 420,000 data points across 75 countries, validated by Harvard researchers, and the platform was acquired by ISG (Nasdaq: III). The top 1% of AI-mature professionals save 31 hours per week. The bottom 50% save six. That's a 5x productivity gap. And 94% of that top 1% use AI multiple times daily, compared to 12% of the bottom half. The gap across our four measured dimensions is even more stark: Creativity 3.7x, Productivity 2.4x, Collaboration 4.3x, and Decision-Making a staggering 13.6x.

This gap isn't closing. It's accelerating. And it maps directly onto labor market reality.

The Numbers That Should Be Front-Page News

In 2025, the U.S. economy grew at a healthy clip. Q2 GDP: 3.8%. Q3: 4.4%. Corporate profits up. Consumer spending stable. By every traditional metric, the economy looked strong.

But payroll employment expanded by only 584,000 jobs for the entire year. The weakest pace since the pandemic year of 2020. Job openings fell to 6.5 million by December, a level not seen since September 2020. Announced hiring plans fell 34% year-over-year to their lowest level since 2010.

Challenger, Gray & Christmas officially attributed 54,836 layoffs to AI. But that counts only the companies that explicitly said "AI" in their press releases. David Shapiro, who has been doing the most rigorous independent analysis of this data through his Labor/Zero research, ran two independent methodologies that converge on the same range.

His first method: excess layoffs analysis. In a non-recessionary year, baseline expectations are roughly 525,000 announced cuts. Actual 2025 announcements: 1,206,374. After controlling for the DOGE federal workforce reduction (293,753 jobs plus downstream effects), tariff-related cuts, technology sector corrections, housing headwinds, and retail restructuring, somewhere between 150,000 and 230,000 layoffs remain unexplained by conventional factors.

His second method: productivity-employment gap analysis. Q3 2025 showed nonfarm business productivity growing at 1.9% annually, with output up 2.8% but hours worked up only 0.9%. That productivity overshoot, against historical baselines, translates to roughly 272,000 jobs worth of reduced labor demand.

Both methods independently land in the 200,000 to 300,000 range. That convergence is the signal. And the mechanism isn't primarily announced layoffs. It's the silent channel: companies deploying AI while simply not replacing departing workers. Seasonal retail hiring in 2025 was the weakest since 2009. The jobs aren't being destroyed in press releases. They're evaporating through attrition.

January 2026 data made it worse: 108,435 announced cuts, the highest January since the financial crisis. Hiring plans hit their lowest January on record.

GDP grew 4% while hundreds of thousands of jobs quietly disappeared. That's jobless growth. It used to be a fringe theory. Now it's in the BLS data, and we don't have a policy framework to deal with it.

The Demand Problem Nobody's Solving

Consumer spending accounts for roughly 70% of U.S. GDP. That spending runs on wages. If wages compress or disappear for large chunks of the population, who buys what the automated systems produce?

I've watched this play out at the company level. When I automated most of my operations, I cut costs substantially. The money that would have gone to twenty-plus salaries stays in the business. My margins are excellent. But those salaries were also someone else's purchasing power. Multiply my situation across every company doing the same thing, and you've got a macroeconomic problem that no individual company has any incentive to solve.

Henry Ford understood this a century ago. He paid his factory workers enough to buy the cars they built. The workers became the market. It was self-interested genius.

Jeff Bezos is making the opposite bet. Amazon is investing $100 billion to automate its fulfillment network. The efficiency gains are staggering. But unlike Ford's strategy, this one doesn't create buyers. It eliminates them.

Shapiro lays out the deflationary death spiral explicitly: Companies automate due to competitive pressure. Laid-off workers reduce spending, default on mortgages, stop paying taxes. Companies face reduced revenue. Production slows. Markets decline. Government becomes insolvent. This happens because the vast majority of household spending is entirely dependent on wages, and we've never had a viable non-human alternative for cognitive production at scale before.

Labor's share of national income has been declining for decades. You can see the trend on FRED, the Federal Reserve's public data tool. Wages have decoupled from productivity output since the early 1970s. AI is accelerating a trend that was already in motion, pushing it past the point where the system self-corrects.

The real risk isn't robots taking jobs. The risk is robots taking jobs while nobody redesigns how income reaches the households that keep the economy running.

Three Income Buckets, and Only One Is Under Attack

The Bureau of Economic Analysis tracks three channels of household income: wages, capital returns, and government transfers. Nearly all the panic focuses on wages, and that panic is justified. But wages are only one of three buckets.

Capital income is doing better than fine. The people who own the robots, the models, the data centers, are accumulating wealth at historically unprecedented rates. Roughly five companies control frontier AI development. Their combined market capitalization tells the story. Returns on capital are growing while returns on labor shrink.

The structural question becomes: how do you move more people from the wages-only bucket into the capital bucket?

This has been solved before, in specific places, by specific mechanisms.

Norway built a sovereign wealth fund worth over $1.6 trillion from oil royalties. Every citizen benefits from the returns. The fund follows the Santiago Principles: separation of administration, government, and oversight. Transparency cuts waste and corruption. It's not socialism. It's fiscal engineering.

Alaska's Permanent Fund sends $1,000 to $2,000 per year to every resident. It was created by a Republican governor. The source is resource royalties. Neither program requires punitive taxation. Both recognize that when resource wealth accrues disproportionately, you need a mechanism to distribute the returns.

Spain's Mondragon Corporation, a network of worker cooperatives, has operated since 1956 with superior firm survival rates compared to traditional companies. Italy's Emilia-Romagna region, built on cooperative structures, is one of the wealthiest regions in Europe. Workers own the enterprise. When automation arrives, the gains flow to the people who did the work.

The same logic applies to AI. If automated systems generate the wealth, humans need an ownership stake in those systems. ESOPs, cooperatives, sovereign wealth funds, baby bonds for every newborn, DAOs, community trusts. These aren't redistribution. They're capital onramps. Ways to give ordinary people a claim on the productivity gains that AI creates.

Shapiro makes the conservative case explicitly: "A government that gives a household a thousand dollars has improved its month. A government that places a thousand dollars in a wealth-generating vehicle has potentially improved its decade." Capitalization can happen through market-friendly mechanisms: auctioning spectrum rights, licensing citizen data, capturing rents on commons. None of this requires punitive corporate taxes. Even the Heritage Foundation is now quoting post-labor economists by name.

The path forward is owning more, not working less.

What Humans Keep, and Why Brazil Should Pay Attention

Two categories of work survive long-term. Both became visible to me only after I'd automated everything else.

The first is the authenticity premium. The barista who knows your name. Live sports where the outcome isn't predetermined. A handmade ceramic bowl with a slight imperfection that proves a human shaped it. A concert where the singer might crack a note. People pay premiums for these things specifically because a human does them. The imperfection is the product.

Shapiro calls this the Quaternary Sector, the Meaning Economy: economically irrational activities where humans irrationally prefer or legally require other humans. Content creators, parasocial relationships, concerts, spa treatments, guided tours. Embodied, local, culturally embedded experiences.

The second is the accountability requirement. Judges, presidents, property owners, military commanders. Legal systems need a human who can be held answerable. You can't put an algorithm in prison. You can't impeach a neural network. Wherever accountability matters, a human stays in the loop. Not because they're better at the task, but because the system demands someone to hold responsible.

Brazil, where I gave this talk, is better positioned for this transition than Silicon Valley. Brazil's culture is the meaning economy incarnate. Music, dance, food, community, carnival, capoeira, street markets, the warmth of human connection baked into daily life. These aren't economic activities to be optimized. They're expressions of human connection that gain value as everything transactional gets automated. Countries that lead in authenticity and human experience will thrive in a post-labor world. Brazil has a head start most countries would envy.

Brazil's Pix payment system is also relevant: an open, permissionless financial infrastructure that removes gatekeepers and increases participation. The kind of financial plumbing that a post-labor economy requires.

But these sectors can't absorb everyone. The Meaning Economy creates real value. It can't replace the scale of wage employment that AI displaces.

The Political Risk Nobody's Naming

This is the part that matters most and gets discussed least.

Throughout modern history, states needed citizens. For soldiers, for tax revenue, for factory production. Citizens had bargaining power because governments couldn't function without them. That mutual dependence produced labor rights, universal suffrage, public education, social safety nets. The state invested in people because people were the productive asset.

Shapiro's most provocative thesis: the period from roughly 1850 to 2050 where workers held meaningful bargaining power may turn out to be the historical anomaly, not the norm. For most of human history, the default arrangement was peasants with zero structural power and elites with all of it. The welfare state, labor rights, unions, these emerged because labor had structural leverage. Elites made concessions because the cost of concessions was lower than the cost of conflict. The Paris Commune, the Russian Revolution, American labor strikes: these "concentrated elite minds wonderfully," as Shapiro puts it.

AI changes that equation fundamentally. When automated systems can handle production, generate tax revenue through corporate productivity, and manage defense, the citizen's bargaining position weakens structurally. Not because anyone is plotting. Because the dependence that created the leverage dissolves.

And here's what makes it worse: automated surveillance, predictive policing, and autonomous security systems could eliminate even the ultimate backstop. Historically, when inequality became extreme enough, the threat of revolution forced concessions. That threat depends on the state being unable to suppress unrest without popular cooperation. That dependency is eroding too.

UBI alone doesn't solve this. Transfers create recipients, not political actors. People dependent primarily on government payments remain vulnerable to political manipulation. Their income can be conditioned, means-tested, or revoked. Capital ownership provides a different kind of security. You can't revoke someone's shares as easily as you can cut a government check.

The window for building ownership structures that include ordinary people is open now. Labor still retains value in many sectors. Democratic institutions still function. Early 20th-century labor movements used their remaining leverage to build institutions, unions, labor law, social insurance, that persisted long after the specific leverage that created them had disappeared. We need to do the same thing, but for capital ownership.

The answer isn't slower AI. The answer is broader ownership before the gains consolidate permanently.

Three Scenarios, One Window

The best case looks like Norway scaled globally. Sovereign wealth funds capitalized through AI company taxes, spectrum auctions, and data licensing. Universal baby bonds that give every newborn a compounding capital stake. ESOP and cooperative structures as the default corporate form. Education that prepares people for ownership and creation, not just employment. This is possible. No new technology is required. Only political will. Timeline: 10-20 years if action begins this decade.

More likely, we get the messy version. A chaotic 10-to-15-year transition where some countries get the economics right and others don't. Mass layoffs in knowledge work (legal, accounting, marketing, customer service) through 2030. Political backlash creates anti-AI movements. Nordic countries and Singapore probably implement capital distribution early. Others lag due to gridlock. Inequality spikes before it stabilizes. Brazil's informal economy (40%+ of the workforce) may paradoxically prove more resilient: these workers are already entrepreneurs, already own their means of production, already embedded in local relationships that AI can't replicate.

The scenario nobody wants to say out loud: winner-take-all dynamics lock in. Five companies and a handful of sovereign states control the AI infrastructure. Citizens lose bargaining power permanently. Something like techno-feudalism, where access to AI-generated abundance is mediated by platform owners who set the terms. Shapiro warns that without structural replacement for labor leverage, societies revert to the historical default: most people with no meaningful structural power whatsoever. Not from malevolence. From structural dynamics. This is the scenario we sleepwalk into if we do nothing.

The difference between these outcomes isn't technological. The AI is coming regardless. The Q3 2025 data already showed the strongest quarterly signal of technology-driven labor substitution: 4.9% annualized productivity growth, output at 5.4%, hours worked at just 0.5%. The difference is whether we build the ownership structures, the capital onramps, and the sovereign funds before the window closes.

What I Tell Business Leaders Who Ask

Stop optimizing for being a better employer. Start thinking about making people owners.

Audit every role in your organization against what AI can already do. Redeploy people toward authenticity and accountability work, the things AI can't replicate. Build internal AI capability deliberately, not as a cost-cutting exercise but as a strategic transformation that includes your workforce in the upside.

What I tell individuals: stop optimizing for being a better employee. Start owning things. Equity, IP, capital, your own business. The employee value proposition is weakening quarter by quarter. The owner value proposition isn't. Build skills in authenticity, the things AI can't replicate: relationships, judgment, cultural fluency, the kind of human connection that Brazil does better than anywhere on earth.

The technology is neutral. The productivity gains are real and accelerating. The question is who captures them.

That question gets answered in the next ten years. And the answer depends entirely on choices we make now, not technologies we build later.

If you run a company, you're already making this choice. Automating without building ownership structures for your workforce is a choice. It just doesn't feel like one yet.

PS: The Messy Middle scenario assumes democratic institutions still function. That assumption is doing a lot of work.

Iwo Szapar is the creator of MemoryOS and the AI Maturity Index (420,000 data points across 75 countries, acquired by ISG/Nasdaq: III). He has worked with 3,000+ companies and 25,000+ professionals at Microsoft, Walmart, and governments worldwide on AI adoption and the future of work. David Shapiro's "Labor/Zero: A Post-Labor Economics Treatise" is available via Kickstarter and informed much of the economic framework in this essay.