A Note from the Future: How AI Changed Lives and the Outcome

Eight years ago I told Startland News the merit-based economy was an illusion ripe for AI disruption. This is what the next fifteen years do to ordinary life, and why the economics make most of it unavoidable.

Written as a dispatch from 2041. Every forward claim is an extrapolation from measurements that existed in September 2026. The seam between the two is marked.

The 2018 position

In April 2018, Startland News profiled me, then founder, Board Chair and CEO of the Kansas City AI Lab, a talent cooperative we built to untangle local technical talent from legacy corporate structures and give practitioners a stake in the work they produced. The headline called the merit-based economy an illusion ripe for AI disruption. The argument underneath it was that machine learning would dissolve the proxies we use to allocate income and status, and that the consequences would land on jobs, poverty and wealth disparity rather than on job titles and robots.

Eight years later the mechanism holds and the target was wrong. The merit illusion did not die. It migrated. AI dissolved the specific proxy in use since roughly 1950, which was demonstrated ability to execute cognitive work, and left the underlying allocation problem completely intact. The disruption also arrived from the wrong end. It came up from the bottom of the ladder rather than down from the top, and almost nobody modeled that correctly. That single detail determines most of what daily life feels like between now and 2041.

Where the data stands, September 2026

Three findings are true simultaneously, and they resist a single narrative. That resistance is why the public argument has been so poor.

The unit cost of cognition is collapsing faster than any input price in industrial history. At constant capability, inference cost has fallen roughly tenfold per year, outpacing compute during the PC revolution and bandwidth during the dotcom buildout. GPT-3 class capability, the generation behind the first ChatGPT release, priced at $60 per million tokens in November 2021. Three years later the same capability was available from a small open-weight model at $0.06, a factor of one thousand. Epoch AI, measuring price at fixed capability milestones, found annual declines ranging from ninefold to nine-hundredfold, with a median near fiftyfold. On September 1, 2026, the Silicon Data LLM Token Expenditure Index, a usage-weighted benchmark of effective market prices, printed 97 cents per million tokens, less than half its summer peak.

Aggregate employment has not moved. Stanford’s Digital Economy Lab, working payroll microdata through June 2026, finds no evidence of widespread economy-wide displacement. Acemoglu’s task-based accounting caps the whole effect at 0.66 percent of total factor productivity over ten years, which is real and small. Goldman Sachs puts near-term displacement risk at roughly 2.5 percent of US employment. At the macro level, nothing is happening.

The entry rung has snapped. Employment of workers aged 22 to 25 in AI-exposed occupations now sits 19 percent below where it would be had it tracked their less-exposed peers. Experienced workers show no comparable gap. The divergence has widened steadily since it was first documented in August 2025. It operates through reduced hiring rather than increased separations, and it concentrates in occupations where AI substitutes for human tasks rather than complementing them. Brynjolfsson and coauthors measured 13 percent on an earlier cut. Hosseini and Lichtinger found adopting firms reducing junior headcount 9 percent relative to non-adopters while senior headcount held flat. In the United Kingdom, exposed job adverts are down 38 percent.

Read together, the shape is unambiguous. The economy has not lost jobs. It has lost the mechanism by which people acquire the right to have one. Entry-level work was never primarily production. It was the apprenticeship toll booth, and the toll booth was the first thing automated.

That is the last measured statement in this piece. Everything below is forecast.

Two results drive everything

If you disagree with the fifteen-year picture, disagree here rather than with the conclusions.

Result one: verification sets the clock

The popular claim is that ideas now move from imagination to reality in seconds. This is true for an identifiable class of artifact and false for the rest, and the boundary is predictable.

Time to artifact decomposes into four terms.

tau = tau_i + tau_s + tau_x + tau_v

ideation, specification, execution, verification

Generative systems drive execution toward zero. That is the entire technological event of the 2020s. Nothing else in the decomposition changed.

Realized speedup is therefore bounded by execution’s original share of the total, and the floor is set by verification, which is not a technological constant. Verification cost is a property of the domain’s feedback loop, and it partitions the economy into three regimes.

• Automated and instant. Code under test, layout, copy, schema transforms, most digital assembly. Verification runs in seconds and can itself be automated. Transformation is total, and ideas genuinely do reach reality in seconds.
• Empirical and slow. Pharmaceuticals, materials, aerospace, civil structures. Verification requires a physical trial that cannot be compressed by intelligence alone. Acceleration is real but bounded, typically 20 to 40 percent of total cycle time rather than orders of magnitude.
• Social and adversarial. Law, governance, clinical medicine at the bedside, institutional trust. Verification is contested by parties with opposing interests, which makes it slow by design. Acceleration is near zero, and several of these domains decelerate, because cheap generation multiplies the volume of claims requiring adjudication without expanding adjudication capacity.

This is why 2041 is uneven rather than uniformly futuristic. A fifteen-year-old in Topeka instantiates in an afternoon what required a funded team in 2018. That same fifteen-year-old cannot instantiate a clinic, a trial, a zoning variance or a reputation. The near-zero marginal cost world is real and it is shaped, and verification cost is the shape.

Result two: the cost of living converges on what could not be automated

Partition household consumption into an automatable sector and a bottleneck sector.

P = w_c × P_c + w_b × P_b

As P_c approaches zero, so does w_c, and P converges on P_b

Cost of living does not converge to zero. It converges to the price of whatever resisted automation. Aghion, Jones and Jones stated the underlying mechanism in 2017: growth is constrained not by capability but by what is essential and hard to improve, and the non-automated tasks remain the weak links that hold the labor share elevated even under near-complete automation.

The honest formulation is therefore narrow. Cost of goods sold for the cognition-intensive component of nearly everything approaches zero. Prices do not, because the binding constraint moved rather than disappeared. By 2041 the binding constraints are land, energy, embodied care, biological time and permission.

Phase I, 2026 to 2031: The Severing

The macro numbers stay healthy and individual experience does not, which makes this the most disorienting five-year stretch in postwar economic life.

The 19 percent entry-level gap widens before it stabilizes, and it spreads outward from software and customer support into paralegal work, junior analysis, junior design, entry-level radiology, bookkeeping and the first rung of professional services generally. Headline unemployment does not spike. Hiring simply stops for people under 28 in exposed fields. Adjustment continues to run through employment rather than base compensation, exactly as the 2026 data showed, so nobody takes a visible pay cut. A cohort just never gets called back.

The dominant condition of this period is illegitimacy without hardship. People are not destitute. They are unanchored. A 24-year-old with a strong degree and no offers is told by every aggregate statistic that the economy is fine, which converts a structural failure into a personal verdict. This is the psychologically worst available configuration, and it is the one we walk into.

Anyone already senior has the strongest five years of their career. The premium on judgment, taste, domain context and willingness to carry liability rises sharply, because verification is the binding constraint and verification requires people who already know what correct output looks like. The 2026 finding that experienced workers in complementary occupations are flat or rising is the leading edge of a much larger repricing.

What replaces the ladder

This is the part that matters most to anyone under thirty reading in 2026, and the part the discourse has almost entirely missed.

The traditional path into data science, analytics and AI strategy ran through a technical entry rung. You were hired to clean data, write queries, build dashboards, maintain pipelines and ship small features under supervision, and after three to five years of that you were trusted with judgment. That path closes, because every task on that list is verification-cheap and therefore fully automated before 2030.

What replaces it is not a different technical entry rung. It is a functional one.

The new prerequisite for practicing data science or AI strategy is roughly a decade spent inside functional, operational or physical industry roles before touching a model. The verifier’s job is to know what correct looks like in a specific business, and that knowledge is not available from a dataset. It comes from having adjusted claims, run a shift, managed a route, priced a book, walked a plant floor, sat in the clinic, or worked a field season. In 2018 domain expertise was the soft complement to technical skill. By 2033 it is the entire job, because the technical skill is free and the domain judgment is the only input that cannot be purchased by the million tokens.

Three consequences follow, and each one is uncomfortable.

• Entry to the profession moves from age 22 to roughly age 32. The technical degree stops functioning as the entry credential and becomes a mid-career addition to operational experience. Bootcamps and masters programs that sell a direct path from classroom to model work stop placing graduates around 2029 and mostly stop existing by 2033.
• The pipeline acquires a ten-year lag that nothing can shorten. A firm that decides in 2034 it needs more verifiers cannot produce them by 2036 at any price, because the input is elapsed operational time. This is the hardest constraint in the 2030s economy and it is entirely self-inflicted.
• The organizations that win the 2030s are the ones that fund apprenticeship as a deliberate cost center starting now. Every firm that harvested training as a free byproduct of junior production loses the byproduct the moment junior production is automated. Training has to be paid for explicitly or it does not happen, and almost nobody budgets for it between 2026 and 2030.

The second-order effect governs everything after. The verification economy runs on senior judgment, and production of senior judgment stops in 2026. That notch propagates for roughly forty years. By 2035 the scarce input in the economy is not compute and not capital. It is people capable of telling whether the output is right.

Phase II, 2031 to 2036: The Repricing

This is the decade in which the convergence result stops being an equation and becomes a household budget.

Software is free. Media is free. Design is free. Legal drafting, tax preparation, tutoring, first-pass diagnostics, translation and most analytical work become ambient and bundled, priced at essentially zero on the margin. The discretionary goods basket deflates year over year in a way that feels like a standard-of-living miracle, and in narrow terms it is one.

Four categories move the other way.

Housing does not get cheaper. The cognitive component of construction collapses, covering design, permitting logic, planning and coordination, but land supply is fixed and zoning is a permission good with an adversarial verification cycle measured in years by deliberate design. Reducing the cost of producing a thing without expanding supply of the fixed input beneath it does not lower the price. It transfers surplus to the owner of the fixed input. This is Ricardo, 1817, and it is the most reliable prediction in this entire piece.

It is also the mechanism by which the transition becomes geographic rather than merely economic. AI income concentrates in a small number of metropolitan areas holding the compute, the capital and the firms. Land scarcity is universal. So the cities capturing the surplus see rents absorb it almost immediately, while the regions capturing none of it still face rising energy and care costs with no offsetting income. The result is a widening gap between places that is larger and more durable than the gap between occupations, and it is the single most predictable political consequence of the 2030s. Kansas City and its peers sit on exactly the wrong side of that line unless ownership is deliberately localized before 2031.

Care does not get cheaper. Childcare, eldercare, nursing, rehabilitation and direct clinical presence are not token streams. Care becomes the largest employment sector in the country and, for the first time in its history, one that pays competitively.

Energy gets more expensive before it gets cheaper. Cognition is nearly free in dollars and remains expensive in joules. The compute buildout collides with generation capacity and transmission permitting, which is another adversarial verification problem, and electricity becomes a first-order term in household cost of living for the first time since the 1970s.

Trust gets substantially more expensive. Once generating a plausible artifact costs nothing, the value of an unverified artifact falls to zero alongside it, and economic weight shifts onto provenance, attestation, audit and custody of record. Cheap generation produces expensive epistemics. By the mid-2030s this sector employs more human judgment than software engineering did at its peak.

The distribution question resolves in this decade in one direction or the other. Acemoglu identified the mechanism in 2024: even where AI raises low-skill productivity it can increase rather than reduce inequality, there is no evidence it reduces labor income inequality, and it is predicted to widen the gap between capital and labor income. Cheap cognition raises the return to owning the means of cognition and lowers the return to being one. No mechanism inside the technology distributes anything.

Phase III, 2036 to 2041: The Settlement

By 2036 the argument is over and the institutions either exist or they do not. Income has substantially detached from employment across the exposed half of the economy through some combination of direct transfers, compute and resource dividends, cooperative equity, and care-sector wages that finally reflect care-sector scarcity. The share of prime-age adults outside formal employment stabilizes well above anything in the historical record.

Aggregate mood then depends almost entirely on a variable that has nothing to do with money.

What a day looks like in 2041

A 31-year-old in Kansas City has never held a conventional job. She holds units in two production cooperatives and gives roughly twenty hours a week to the one she cares about, a small manufacturing operation where her actual function is deciding what is good, because machines handle everything downstream of that judgment. Her income is roughly half distribution and half cooperative yield. Housing consumes more than half of it. She would describe her life as busy and she would be correct.

A 58-year-old former systems architect spends most of his week on verification. He signs off on output, carries liability and is the person whose name goes on the record. This is the best-compensated work in the economy and there are not enough people qualified to do it, because the pipeline closed in 2026. He is exhausted, wealthy and quietly angry about the notch.

A 24-year-old is the variable the entire forecast turns on, and he exists in one of two versions.

In the first, his city built the institutions. He belongs to a restoration guild that meets four mornings a week in a building the city gave them for a dollar. There is a foreman who expects him at seven. There are six other people who notice whether he shows up and who give him a hard time when he does not. He is eleven months into learning to hang a door properly and he is still not good at it, which is the point, because the wood does not care about his feelings and cannot be prompted into compliance. On Thursdays he plays in a league. He is broke in the sense that his housing costs more than half his income, and he is not lost.

In the second version, his city built nothing. He receives the same money on the same day of the month. He wakes around eleven because there is no reason to wake earlier. The first thing he touches is a screen and the last thing he touches is a screen, and in between the screen has been engineered by systems far more capable than he is to hold him exactly at the threshold of engagement without ever requiring anything of him. Nothing he encounters resists him. Every difficulty is optional and therefore not a difficulty. He has no foreman, no crew, no Thursday, no skill he is eleven months into and bad at. He speaks aloud to another human being perhaps twice a week. He is not poor. He is not hungry. He is twenty-four years old and nothing in his life is load-bearing, and when he is asked what he does, he has no sentence.

The material conditions of those two lives are nearly identical. The subjective conditions are not comparable. That gap, not GDP and not token prices, is the outcome the title refers to.

In aggregate, the hours redistribute into six modes. Chosen difficulty, meaning instruments, kilns, climbing, restoration and long-form mathematics, because when execution is free the deliberate reintroduction of constraint becomes the point, in the same way people took up marathons after cars were invented. Care, which is the largest single reallocation of human hours in the period. Inquiry, meaning amateur science, local history, genealogy and independent research conducted with a research assistant that costs less per month than coffee. Competition, because status is positional by definition and its supply is fixed at 100 percent of the distribution regardless of productivity, so competitive intensity rises as material scarcity falls. Stewardship of land, buildings, archives and neighbors, because maintenance becomes a vocation once production stops being one. Assembly and ritual, the resurgence least anticipated by the technology sector and most predictable from the psychological literature.

There is a seventh mode, described above. Where institutions were not built, people did not flourish. They drifted, and the losses concentrated among young men in low-density regions.

The five functions problem

The psychological evidence for all of the above already exists and has since 1933.

Marie Jahoda’s latent deprivation model holds that employment supplies a manifest function, which is income, plus five latent functions: time structure, social contact, status, collective purpose and activity. Meta-analysis confirms that employed people score higher on all five latent functions and on the manifest one. Longitudinal meta-analyses establish that unemployment causally reduces well-being and mental health, with effects worsening the longer it persists.

Taken at face value, that is fatal to post-labor optimism. The reframing sits in the German population data, which found that people out of the labor force, meaning students, homemakers and retirees, report significantly fewer latent benefits than the employed but significantly more than the unemployed, and that on identity and status specifically there is no significant difference between the employed and the out-of-labor-force group. Only the unemployed report lower status than everyone else.

Unemployment and post-employment are distinct states with opposite outcomes at identical hours. The retiree and the laid-off analyst share a calendar and share nothing else. The damage is not produced by the absence of a job. It is produced by the absence of a job one was supposed to have, combined with the absence of substitute institutions supplying the five functions.

This converts the philosophical question into an engineering specification. A functioning 2041 life requires an institution that supplies externally anchored time structure, effortful activity against real resistance, regular and non-optional contact with people one did not select, participation in a project larger than oneself, and legible recognition from a community whose judgment one respects. Employment bundled all five and charged a waking life for the package. Nothing else bundles all five by default. That bundle, rather than the income, is the hard problem of the transition.

The best available evidence on behavior at the margin is the OpenResearch Unconditional Cash Study, which enrolled 3,000 participants across Illinois and Texas, gave 1,000 of them $1,000 monthly for three years against a control group of 2,000 receiving $50, and represented a 40 percent income increase for recipients averaging under $29,000. Recipients worked 1.3 fewer hours per week and were 10 percent more likely to be searching for work. The authors attributed the reduction to a high marginal valuation of leisure. Alongside it, recipients spent more on food, transportation and housing, gave more financial assistance to others, showed greater agency in goal setting, pursued work they found interesting or meaningful, moved into better housing, and enrolled in school or training at higher rates in the final year of the pilot.

Every item on that list describes a person recovering the capacity to choose. The public argument consumed two years over fifteen minutes per day and filed the meaningful-work finding under laziness. The objection to distribution was never empirical. It was a moral commitment to employment as the legitimate basis for a claim on resources, and that commitment is the thing that breaks between 2031 and 2036.

Three failure modes

Rent capture. Cognition goes free, land and permission do not, and the surplus lands entirely on fixed-supply assets. This is the default path and requires no coordination, only continued inaction on land use, licensure and energy permitting. It is the most probable adverse outcome and the least dramatic, which is precisely why it receives the least attention.

The severed ladder. Adjustment runs through reduced hiring of the young rather than separations, so a cohort that never apprentices never becomes senior. The result is a forty-year expertise notch that arrives exactly when the verification economy needs senior judgment most.

Legitimacy collapse. If income detaches from employment while the moral framework still treats employment as the source of desert, the result is a population that is materially provided for and socially delegitimized. That combination does not produce contentment. The World Economic Forum’s 2025 employer survey projected 92 million jobs displaced and 170 million created by 2030, a net gain of 78 million. Net gains offer no consolation when the displaced and the created are different people in different places.

Forecast summary

Stated so they can be checked.

1. By 2029, the entry-level employment gap in AI-exposed occupations exceeds 25 percent and extends to at least six additional occupational families beyond software and support.
2. By 2031, median compensation for senior verification and liability-bearing roles rises at least 60 percent in real terms relative to 2026.
3. By 2033, housing and energy together exceed 55 percent of median household expenditure in metropolitan areas, driven by falling denominators rather than rising rents alone.
4. By 2034, care work exceeds retail and transportation combined as a share of employment, and real wages in the sector rise faster than any other major category.
5. By 2035, the median age of entry into professional data and AI roles exceeds 30, and operational industry experience displaces the technical degree as the primary hiring signal.
6. By 2036, at least four US states operate a substantial non-employment income floor, whether financed by resource dividends, compute levies or direct transfer.
7. By 2038, provenance and verification services constitute more than 3 percent of GDP.
8. By 2041, the strongest predictor of self-reported well-being among adults outside formal employment is membership in an institution supplying three or more of Jahoda’s five latent functions, and it outperforms income as a predictor.

The last of these is the one worth building policy around.

What the 2018 thesis got wrong

The 2018 claim was that the merit-based economy was an illusion ripe for AI disruption. The sentence was correct and the emphasis was misplaced. The illusion was never merit itself. It was the belief that employment functioned as a measurement device, reading off human worth and paying accordingly.

Employment was never a measurement device. It was a rationing device, allocating access to the material basis of a life through a proxy, hours of cognitive output, that happened to be scarce. AI did not reveal the proxy as unfair. It made the proxy abundant. A rationing device built on an abundant input does not become fair. It becomes non-functional, and the fifteen years that follow are spent discovering that no replacement was ever designed.

The cooperative structures being built in Kansas City in 2016 were not a community initiative. They were a structurally coherent response to a world in which the return to hours falls toward zero and the return to ownership rises without bound. The correct policy response to that world is not retraining. It is ownership transferred before the transition completes, and the window for doing it cheaply closes around 2031.

The technology question was settled early and it was the easy one. Ninety-seven cents per million tokens settled it. The open questions were always about who holds the compute, who holds the land, who holds the permission, and what supplies structure, contact, purpose and status once the job no longer does. Those questions were answerable in 2018. They remain answerable now. They stop being answerable cheaply in about five years.

About the Author

Brian Curry is a Kansas City-based Machine Learning & AI Strategist, Marketing Economist, and Applied Researcher. He drives Vector1, an independent research studio working at the frontiers of cognitive AI, causal inference, marketing economics, and knowledge engineering to build simulation-driven intelligence systems.

Across more than twenty years, Brian has led advanced data science and analytics initiatives at enterprises ranging from $4B to $125B in revenue, including Koch Industries, Tractor Supply, Vail Resorts, Hallmark, Garmin, and AT&T. His work systematically applies economic thinking and causal inference to connect marketing investment to structural outcomes, operational decisions to economic value, and AI adoption to permanent competitive advantage.

He is also the founder of KC AI Lab, a 1,500-member AI/ML education community in the Midwest.

Key Research & Open-Source Projects

GitHub: @Bodhi8

Causal Inference & Marketing Economics

• PyCausalSim: A Python framework for simulation-driven causal discovery and evaluation, mapping counterfactual scenarios to isolate true structural drivers.
• Papilon: A complex-system simulation and optimization library optimized for Media Mix Modeling (MMM) and adaptive architectures.
• Daedalus: A causal analytics framework designed for money-flow mapping and macro-economic tracking.

Algorithmic Engineering & Applied Forecasting

• Regimecast: Applied research and tooling for time-series regime detection, structural breaks, and economic forecasting.
• CarryNet: Open-source research and financial modeling focused on commodity carry trade analysis.
• Foodmesh-ml: A machine learning, optimization, and simulation framework engineered for modeling distributed food systems.

Cognitive AI & Semantic Knowledge Architecture

• Memory-Node Encapsulation (MNE): A cognitive AI architecture and novel data structure for artificial episodic memory (NeoCortex-M).
• The Autonomy Economy: A macroeconomic framework projecting the dual impacts of massive productivity growth and extreme structural price deflation in an AI-automated world.
• MeaningFlow: A semantic intelligence framework focused on content modeling, embeddings, clustering, and automated coverage gap analysis.

Contact: brian@vector1.ai | Vector1 | GitHub | LinkedIn

This article may be shared with attribution under Creative Commons Attribution 4.0 International (CC BY 4.0).

References

1. Startland News (2018). Merit-based economy is an illusion ripe for AI disruption, talent coop founder says. startlandnews.com/2018/04/ai-disruption/
2. Brynjolfsson, E., Chandar, B., Chen, R. (2026). Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence. Stanford Digital Economy Lab.
3. Acemoglu, D. (2024). The Simple Macroeconomics of AI. NBER Working Paper 32487.
4. Aghion, P., Jones, B. F., Jones, C. I. (2017). Artificial Intelligence and Economic Growth. NBER Working Paper 23928.
5. Appenzeller, G. (2024). Welcome to LLMflation. Andreessen Horowitz.
6. Silicon Data LLM Token Expenditure Index (SDLLMTK), September 2026.
7. Epoch AI, benchmark-level pricing analysis, 2026.
8. Vivalt, E., Rhodes, E., Bartik, A., Broockman, D., Miller, S. The Employment Effects of a Guaranteed Income. OpenResearch Unconditional Cash Study.
9. Hosseini, A., Lichtinger, G. (2025), and related displacement estimates surveyed in AI and Jobs: A Review of Theory, Estimates and Evidence (2025).
10. Jahoda, M. (1982). Employment and Unemployment: A Social-Psychological Analysis. Cambridge University Press.
11. Paul, K. I., et al. (2023). Employment status, psychological needs, and mental health: meta-analytic findings concerning the latent deprivation model. Frontiers in Psychology.
12. Paul, K. I., Batinic, B. (2010). The need for work: Jahoda’s latent functions of employment in a representative sample of the German population.
13. World Economic Forum (2025). Future of Jobs Report 2025.
14. Baumol, W. J. (1967). Macroeconomics of Unbalanced Growth: The Anatomy of Urban Crisis. American Economic Review 57(3).
15. Keynes, J. M. (1930). Economic Possibilities for our Grandchildren.