The displacement is real. I want to start there because the optimist case is only credible if it begins by staring directly at the pessimist case, and sophisticated audiences can smell hand-waving.

Customer service teams are shrinking. Legal research that required junior associates is now performed in seconds. Back-office functions are being automated across every industry simultaneously. And the breadth is what makes this different from previous waves. This is not manufacturing jobs moving offshore or secretaries replaced by email. AI is rolling across blue collar, white collar, professional, creative, and public sector employment at the same time. That simultaneity is historically unprecedented.

The depression case writes itself. Employment is income and income is consumption. Consumption is roughly 60% of GDP in Australia, closer to 70% in the United States. If you displace workers across every tier you lose not just jobs but the spending those jobs funded. Fewer restaurant meals, fewer car purchases, fewer mortgages being serviced. Businesses that serve those consumers contract, laying off more workers, which reduces consumption further. If you take the job-loss categories at face value and assume nothing else changes, what you are describing is structurally a depression.

That analysis is real as far as it goes. But it stops at first-order effects and declares the game over. The assumption that nothing else changes is the fatal error. Everything else changes.

The historical proof

In 1800, roughly 80–90% of the Western workforce was employed in agriculture. Mechanisation destroyed those jobs systematically over 150 years. By any first-order analysis, this should have produced permanent mass unemployment of catastrophic proportions. Ninety percent of all jobs, gone.

It did not produce catastrophe. And, importantly, there was no single replacement industry. No government programme created “the service economy.” Instead, hundreds of new occupations and entire sectors emerged: manufacturing, retail, professional services, entertainment, healthcare, education, financial services, tourism, telecommunications. Most would have been incomprehensible to a farmer in 1820. Nobody planned this. It emerged organically from cheaper food, freed-up labour, surplus capital, and human ingenuity finding new wants to serve.

The people who lived through the early phases experienced genuine hardship. The Luddites were not irrational. But employment today stands at roughly 95% despite 90% of 1800’s jobs having been eliminated. The solution was not a silver bullet. It was a hundred 1% effects that nobody could have predicted.

The historical base rate for technology-driven displacement leading to permanent economic contraction is zero. Every general-purpose technology – agriculture, steam, electricity, the automobile, the computer, the internet – has followed the same pattern: short-term disruption, medium-term adaptation, long-term prosperity that dwarfs what came before.

The productivity reframing

As a mathematician, I find it useful to think about what AI replacing labour actually is. At its core, it is a productivity improvement. First, the cost of performing a task drops, in some cases by 90% or more. Then, the savings distribute across three parties, and tracing those flows is more illuminating than counting lost jobs.

The share that accrues to capital gets reinvested. Critically, when the cost of building a company drops from $2 million to $200,000, the number of fundable ideas explodes. I can see this directly from an angel investing vantage point. The pipeline of investable early-stage companies does not shrink in an AI-driven economy. It booms.

The share that flows to customers increases real purchasing power. If legal services cost 90% less, healthcare administration 80% less, and logistics 70% less, effective household income rises substantially even if nominal wages are flat. Services previously accessible only to the wealthy become universal. Think air travel, telecommunications, and computing.

The share captured by remaining labour is the least intuitive. When AI makes a worker dramatically more productive, that worker becomes more valuable, not less. A lawyer who can do the work of ten lawyers does not get paid one-tenth of the old rate. They capture a premium, because judgment and client relationships are the scarce input. One architect with AI does what a firm of twenty did. That architect does not earn one-twentieth of the old firm’s revenue.

And consider the framing. “Unemployment triples from 5% to 15%” sounds catastrophic. “Employment remains at 85–90%” tells a different story. An economy where 85% of people are employed, each producing dramatically more output, with falling costs across the board, is not a depression. It is a boom with a distributional challenge, and though it’s serious problem, it’s one humanity has solved repeatedly.

The immediate absorbers

Some of the 1% effects are prosaic, which is partly why they get overlooked. Australia currently has 300,000 to 400,000 unfilled job vacancies, concentrated in healthcare, construction, education, and hospitality. These are physical-presence, human-relationship jobs that AI cannot replace in the short or medium term. The first tranche of displaced workers is entering a labour market with enormous unmet demand in sectors that are the hardest for AI to touch. And if AI simultaneously reduces administrative costs in those sectors such as rostering, compliance documentation, and medication management, the savings flow to human carers and tradespeople who are now the scarce, valued input. The retraining gap between “insurance claims processor” and “aged care coordinator” is dramatically smaller than the farmer-to-factory-worker transition that the agricultural revolution managed.

Then there is the transition itself. Every major organisational transformation generates its own industry of advisors, implementers, and trainers. AI is no different, except in scale. Think strategy consultants, AI agent builders, integration engineers, data specialists, change management professionals, ongoing maintenance. Multiply this across every bank, insurer, hospital network, university, and government department in the economy and the transition becomes one of the largest employment categories of the decade. And it is not a one-off. AI systems evolve rapidly. The organisations that adopt in 2026 will need to upgrade and restructure again in 2028 and 2030. The transition industry is permanent.

Cathedral economics

When economies generate surplus capital, some portion gets directed toward ambitious, labour-intensive projects that generate value across centuries. The Pharaohs built pyramids that employ tens of thousands of Egyptians in tourism 4,500 years later. The medieval Church built cathedrals that underwrite entire regional economies across Europe today. The pattern repeats: Versailles, the Alhambra, the Forbidden City. Sydney Harbour’s mansions are the miniature contemporary version. They are perpetual employment machines for architects, stonemasons, landscape designers, and artisans.

Modern engineering continued at greater scale. The Empire State Building employed 3,400 workers daily. The Channel Tunnel engaged 13,000 over six years. The semiconductor fabs TSMC and Intel are building today, each costing tens of billions, are the latest cathedrals, requiring artisan-level precision in cleanroom construction and ultra-fine engineering.

If AI generates the capital surplus the productivity thesis suggests, the question is not whether cathedrals get built but how ambitious. Space infrastructure, ocean habitats, desert greening, and city-scale reinvention are all massively labour-intensive, and each generate value across generational timeframes.

And cathedrals do not just employ labour. They employ artisans – masters of craft. AI handling the routine while surplus capital flows toward ambitious physical projects creates a new premium on irreducibly human, mastery-based work. It funded the Renaissance. I suspect it funds whatever comes next.

The IRL economy

There is a complementary force that works from the demand side rather than the capital side. As AI mediates more of working life (fewer meetings, fewer colleagues, less incidental social fabric) the human need for physical presence and shared experience intensifies.

Consider hiking. Healthy, cheap, endlessly varied, available in Australia’s world-class national parks. Yet when I walk the trails, I regularly see almost no one. It is not hard to see participation doubling or tripling. If it does, the downstream employment ripples in every direction: trail maintenance, ranger services, gear retail and manufacturing, regional hospitality, transport, guided experiences. Someone needs to make and sell the hiking boots.

Hiking sits on a long spectrum. The broader IRL economy encompasses fitness and movement, community spaces, sport and recreation, live performance, the outdoor economy, and learning as recreation. Examples include pottery classes, language schools, musical instruments, woodworking, and astronomy clubs. Every one of these is irreducibly human, physically local, and scales with available leisure time and disposable income. Both increase as AI-driven productivity gains work through the economy.

There is something else worth noting. Many of the jobs being displaced are jobs nobody finds particularly fulfilling. Data entry, insurance claims processing, routine compliance reporting. The IRL jobs that replace them tend to carry higher intrinsic meaning. The transition is not just from employment to employment. It is potentially from low-meaning work to high-meaning work.

More 1% effects

The ones above are the ones I find most structurally interesting, but once you start looking, the list grows fast.

The longevity economy. If AI accelerates drug discovery, diagnostics, and personalised medicine – and the early evidence suggests it will – people live longer and healthier. Extended healthspan means extended working lives, extended consumption, and an enormous care and wellness economy that barely exists today. Japan is the leading indicator. Australia is twenty years behind it.

The trust economy. When AI can generate anything, from text to images to video and credentials, the premium on verified authenticity rises. Human-verified journalism, authenticated provenance for goods, identity verification, audit and assurance services. The more AI produces, the more someone needs to confirm what is real. Accountants and auditors may find that AI eliminates their routine work but dramatically expands their assurance role.

The personalisation economy. Mass production gave us cheap, identical goods. AI-driven manufacturing gives us cheap, customised goods such as shoes fitted to your gait, nutrition plans calibrated to your bloodwork, and education pathways shaped to how you actually learn. Each layer of personalisation creates work: in design, in data, in the human judgment that sits between the algorithm and the customer.

The sovereignty economy. Governments are re-shoring critical capabilities such as semiconductors, pharmaceuticals, energy, and defence, and and AI accelerates this by making smaller-scale production viable. Every country that decides it needs domestic capability in something it previously imported is creating jobs that did not exist locally before.

And then there are the ones nobody can name yet. In 1995, if you had described social media managers, app developers, YouTubers, UX designers, and podcast producers, you would have been escorted from the room. The most important 1% effects in the AI transition will be categories of work that sound equally absurd today.

The hundred 1% effects

Roy Amara gave us the corrective lens: we tend to overestimate the effect of a technology in the short run and underestimate the effect in the long run. Applied to AI and employment, this works in both directions simultaneously. In the short run, the displacement will be slower than headlines suggest. Regulation, institutional inertia, union resistance, implementation complexity, and sheer organisational incompetence will drag the timeline out. Autonomous trucks will not replace every driver in 2027. The depression scenario assumes a step-function shock. The reality will be a rolling wave spread across a decade or more.

But in the long run, the offsetting forces that create new employment, new industries, and new sources of demand will be far larger than anyone currently models. And this transition has a structural advantage over every previous one. The internet is a powerful idea replication engine. When someone invents a new business model or job category, it is visible worldwide almost immediately and can be copied, adapted, and scaled within months. In the agricultural transition, new job categories took generations to propagate. In the AI transition, the lag between “jobs destroyed” and “new jobs discovered and scaled” compresses from generations to years.

Stacked together, the effects form a self-reinforcing loop rather than a death spiral. AI-driven productivity generates surplus. Surplus distributes across capital, customers, and remaining labour. Capital reinvests and funds new company formation at unprecedented scale. Customers benefit from lower prices and spend into the IRL economy. Remaining workers earn premiums as the scarce human input. Surplus funds cathedrals. New job categories propagate globally at internet speed. None of these is sufficient on its own. That is the point. The solution is a hundred 1% effects, compounding over time, driven by the same forces that turned agricultural mechanisation into the greatest period of prosperity in human history.

The angel lens

As an angel investor, I find the present moment not threatening but extraordinarily compelling. As I explored in The Stagflation Paradox, the best angel vintages tend to come from periods of disruption when the opportunity set expands, valuations are disciplined, and founders are solving problems the market hasn’t fully recognised yet.

The companies that will be built in the next twenty years, serving needs we cannot yet articulate, in industries that do not yet have names, will make the current economy look as quaint as an 1820 farm looks to us today. The pattern recognition I described in The Walking Investor keeps surfacing the same signal: when the cost of doing something drops by an order of magnitude, the number of things worth doing explodes. The angel investor’s job is to be in the room when those ideas arrive.

Amara was right. He usually is.

Richard Moore is co-founder of MooCoo Ventures, an angel syndicate that co-invests alongside Brisbane Angels, one of Australia’s most active angel groups. He has made over eighty personal angel investments since 2013.

moocoo.vc