The Valley PaperSources1994

The Valley Between Here and Abundance

AI will make the professional middle class’s work cheap years before it makes their lives cheap. This paper is about those years.

Version 1.0, October 28, 2026. Sources

By Brian Senjem, Pine, Colorado

Before the argument starts

Where I stand. I want abundance done right, and I support building and using superintelligence responsibly. I do not think those are two positions. I have followed Peter Diamandis since I read his book Abundance in 2012, and I still believe the destination he describes is real and worth building. I also think the race to get there first, the wholesale scaling and the staggering sums behind it, is irresponsible, and that the other extreme, the view that it all ends in catastrophe, is no more useful, because it names a risk and hands nobody anything to do. I am writing from the middle, about the years in between. The abundance on the stage and the abundance in the fine print are two different things, and the people who carry this economy fall between them. This paper names that gap. I first wrote about this kind of problem in 1994, in an MBA paper called “Job Security: The Age of Anxiety,” when white-collar workers had just joined blue-collar workers in fearing for their jobs. Thirty-two years later, the fear has a mechanism.

How this paper was written. I wrote it with extensive help from Claude, Anthropic’s AI, and I say so plainly. Working with superintelligence on this problem is the point of the paper, so the method matches the message. I supplied thirty years of running businesses and doing the books for a few hundred more. Claude supplied research reach, organization and the willingness to tell me when a number was wrong.

What this paper is, and is not. It is a diagnosis. It names a problem, shows it from eighteen directions, and lists possible solutions for other people to test, build on or replace. It does not argue for any one of them. It leaves gaps on purpose so that others can fill them. Every number in it has a source, and the list is posted beside it.

Part 1: Two sentences, a week apart

Peter Diamandis is the most-read voice for the abundant future. He is a physician and engineer who founded the XPRIZE Foundation, co-founded Singularity University, and wrote Abundance in 2012. He runs the Abundance360 summit, the Moonshots podcast and the Metatrends newsletter, and his view of AI is unreservedly optimistic. This paper quotes him because he has written more, and more clearly, about where this is going than anyone else, and because I have followed him from the start. When I say the abundance community, I mean the people who follow his work.

On August 26, 2026, Peter Diamandis published a piece called The Death of the Company. Citing Salim Ismail’s pilots with ten companies, he wrote that “AI-native organizations will be approximately 100x more performant than today’s companies,” and “you will do it with about 20% of your current headcount.” His estimate for middle management: it “collapses by 60 to 80%.”

On September 2, he published There Is No Job Apocalypse. It cites the World Economic Forum: 170 million jobs created and 92 million displaced by 2030, “a net gain of 78 million jobs.”

I believe both of them. The gap between those two sentences is where Linda lives. Linda is a 52-year-old CPA with a four-person practice in Denver, and you will meet her properly in Part 2. For now, know that her firm sells exactly the work the first sentence describes, and that the second sentence has no row for her.

How he closes the gap, and why it stays open. His bridge is that entrepreneurs will spawn “five times as many startups,” companies “four to five times more numerous but 75-80% smaller.” The trouble is that the new companies are built on the same design as the one that shed the people: a hundred times the output on a fifth of the staff. Only 6.9% of new business applications this year plan to pay anyone a wage at all. Five times more companies at zero payroll is zero payroll. Part 3 makes the full case.

He has already named the valley. This is the part I did not expect to find. Five months before those two pieces, in an essay called From UBI to UHI (In 3 Steps), under the heading “THE VALLEY THAT MUST BE NAMED,” Diamandis wrote: “Name the valley. It runs from approximately 2026 to 2031.” He called it “a design problem,” and he asked whether the transition could bridge “the valley between here and there.”

I am taking him at his word and at his dates. Where I part from him is on who is standing in the valley and what the bridge has to carry.

His valley holds “the logistics managers, the paralegals, the customer service workers, the junior analysts,” and “the 57% of Americans living paycheck to paycheck.” His bridge is a replacement income of $3,000 a month, “$36,000 per year,” followed by an automation dividend. Those are serious proposals for the people he names, and I am not here to argue against them.

But a $36,000 check does not replace Linda’s $235,000. It does not touch her mortgage. It does not pay the taxes her income paid, or support the two households that lean on hers. The people who carry this economy are not in his valley, and his bridge was not built for their load. He named the valley. He did not count everyone in it.

What the abundance case gets right, and what it promises. AI has collapsed real barriers. I watched it happen inside Easytrack, my previous company: work that took a trained person an afternoon started taking a prompt and a review. The floor Diamandis describes is real and worth building. His near-term target is a bundle of basic human needs for a family of four at “$1,000 per month or less,” by 2027 to 2029. On Moonshots #298 that floor was described, in so many words, as basic subsistence. Above that floor sits the $3,000 replacement income. In January, describing a plan at about that level, he wrote: “To be clear, that’s not survival-level subsistence. It’s lower-middle-class security. For doing nothing.”

Then there is the ceiling, the version that gets sold from the stage: universal high income, every disease cured, unlimited energy, anything you want printed on demand. In his own phases, that is Phase 3, dated 2031 to 2035.

So there are two abundances. One is a floor of basic needs and lower-middle-class security, arriving first. The other is the Star Trek version, arriving last. Lower-middle-class security is below where Linda already lives, and below where most of the country lives. In 2025, about 22% of households had income under $40,000, so a $36,000 replacement income is a raise for one household in five and a pay cut for the other four. For the carrying class it is a 67% to 86% pay cut, and that is before the debt. A $1,000-a-month basket of basic needs is the standard of living of households under $15,000 a year, about 7% of the country. The ceiling is a decade off. The people I am writing about fall between the two.

The floor also leaves one thing out, and it is the thing the middle class was built on. The promise was never “your needs will be met.” The promise was “you can climb.” A floor does not answer that, and I come back to it in Part 3.

The thesis in one paragraph. The map shows the destination, and its author has now marked a valley on it. What the map does not show is who is standing in the valley, what they owe, and what they carry. This paper names them.

A note on tone. This is not an attack on Peter Diamandis. It holds him to his own definition and his own dates. His pieces are the best statement of the destination anyone has written, and his March essay is the clearest admission of a valley from the abundance side. That is why this paper quotes them.

Part 2: The problem, named

The name

AI will make the professional middle class’s work cheap years before it makes their lives cheap, and the Valley is the name for those years.

The Valley is also:

The timeline

Displacement is measurable now, in 2026. Relief has dates too, and they are Diamandis’s own. Phase 1, “The Fracture & the Floor,” runs 2025 to 2028, when “mass displacement arrives faster than safety nets can respond.” Phase 2, “The Automation Dividend,” runs 2028 to 2031. Phase 3, “The Great Deflation & the UHI Moment,” runs 2031 to 2035; that is when AI and robotics collapse the cost of transportation, housing, health care, food and energy.

Read that again from Linda’s side of the table. The costs she pays start falling in 2031, and the fall runs to 2035. My own judgment is that 2031 is the beginning of relief, not the arrival of it, and that relief adequate to her bills comes around 2034 to 2036. That is eight to ten years from today, which is why I call it a five-to-ten-year valley. I am stating that as my estimate laid on top of his dates, not as a forecast I can source.

He set a shorter window himself, in September: “The founders who get hurt in the next five years will be the ones who still believe they have a wall.”

The technology is exponential and the institutions are linear. In 1994 I recommended short-time compensation in every state. Thirty-two years later, 31 states have it on the books and few employers use it. The 1994 fear passed because there was a growing sector for displaced people to move into. Part 3 shows why that exit is the one closing now.

Who is in it: the carrying class

The middle class is not only defined by what you make. It is also defined by who depends on you.

Four traits put a household in the carrying class:

  1. Income between the 60th and 90th percentiles of households. In 2025 that is about $110,000 to $261,000. The percentile is the definition; the dollars are dated and move every year.
  2. Income from labor. For the average household in every quintile, labor income is the bulk of income (Congressional Budget Office, 2023 data); for the middle quintile it is 73% (Treasury, 2025 income levels). Capital is a larger share only at the very top.
  3. Thin assets outside of the home they own.
  4. No power to set the price of its own work. The employee takes the raise she is given. The small-firm owner raises prices until a client leaves. Nobody in this class reprices their own labor.

How big it is

About 41 million households sit between the 60th and 90th percentiles. That is 30% of the country’s 137 million households.

Also at risk: about 7 million more households from the 90th to the 95th percentile, about $261,000 to $354,000. If their income goes, they are in the same position unless they hold real assets, and many of them carry bigger houses, bigger mortgages and more commitments instead. My own judgment is that only households with real assets outside the home, and roughly above $400,000 of income, are clear of this. That $400,000 is my line, not a Census one.

Behind those households sit 36.2 million small businesses (SBA, 2026), about 93 million adults with a bachelor’s degree or higher (BLS, 2025), and the 19% of American workers, about 28 million people, whose jobs Pew rates as most exposed to AI (2022 data).

Why it is the load-bearing wall

Take that wall out and you do not get a smaller economy. You get a different one.

Linda

Linda is a 52-year-old CPA with a four-person practice in Denver and 150 clients. She is based on a real CPA, with details changed for her privacy.

Her husband was the main earner until a job-site injury in 2015 left him working part-time. Her practice paid her about $285,000 a year at its peak, from 2019 to 2022, and those best years went to two college tuitions; it was the only stretch when the household sat above the 90th percentile. It pays about $235,000 now, and with her husband’s part-time income the household is around $250,000. Her fees have been falling for years while her costs keep rising, and she has held on by adding clients and services. That is how she got to 150 clients.

She has 15 years to go until full retirement age, which Social Security sets at 67 for anyone born in 1960 or later. She is not about to retire, and the government does not expect her to. She only recently started saving for retirement, because until then she was putting her kids through college. That is why people like Linda keep working well after the kids are out of the house. College costs that much.

Her balance sheet: about $340,000 in retirement savings; a house bought in 2013 for $475,000 and now worth about $1.1 million; a $280,000 mortgage, refinanced in 2021 at 3%, at about $1,180 a month in principal and interest; and about $30,000 of Parent PLUS loans from the kids’ college, at 5.3 to 7.5% depending on the year, about $340 a month. Most of her net worth is the house.

She already uses the AI tools. She is good with them. And she is doing the math.

I put her in Denver on purpose, so that no reader from the abundance side can wave her off as rural.

Mike

Mike is a 44-year-old licensed project architect at a 40-person Denver firm, earning about $120,000. He is a composite. His wife is a nurse earning about $100,000, so the household is about $220,000. Their kids are 12 and 15, so college is still ahead. They bought their house in 2022, at peak prices, with a mortgage sized for two incomes, and their retirement savings are thin. He has 23 years to go until 67.

His firm adopts AI drafting and design tools, and over two years it goes from 40 people to 22, with project architects cut alongside the drafters. Without his income the household falls to about $100,000, just below where the carrying class starts, carrying a mortgage sized for $220,000.

His profession has been through this once. Employment at architecture firms fell by more than a quarter between 2007 and 2011, while total U.S. employment fell about 4% between the same two years, and firm revenue fell 40% from 2008 to 2011 (BLS, AIA). The firm’s cuts in this paper are a scenario, not a reported case. The last time is not.

Mike is worse placed than Linda in three ways. He is an employee, so one decision by one employer ends his income. He has no practice to sell and no clients to keep. And college is in front of him, not behind.

What this problem is not

Part 3: The problem from eighteen directions

The same problem looks different depending on where you stand. Here it is from eighteen places, in four groups: how the hit arrives at the firm, why the household cannot absorb it, why it spreads through the economy, and why reaching abundance does not fix it. Each passage is short. Each one is built the same way: the claim, how it works, one number you can check, and one line to carry away. The codes (A1, B3) are there so you can refer to a passage by name.

A. How the hit arrives at the firm

A1. Tasks, jobs and the P&L.

“AI takes tasks, not jobs.” You have heard this line, and it is true as far as it goes. It protects a job description. It says nothing about a firm’s revenue.

Linda’s firm sells the tasks. Bookkeeping, reconciliations, sales-tax filings, payroll returns, the first draft of a tax return: those are the tasks, and they are the invoice. For anyone who bills for tasks, task displacement is revenue displacement. The job description survives; the fee does not.

There is now a measurement for this. In October 2025, a group of 33 researchers, among them Max Tegmark, who wants superintelligence banned until it can be proven safe, scored GPT-5 at 57% of a well-educated adult across ten kinds of thinking. It scored near the top on knowledge, reading, writing and math, and zero on remembering anything from one day to the next. The half it has is the half Linda bills for. The part it lacks is the part she has never put on an invoice.

For a professional worker like Mike, the same line means something slightly different. The firm keeps the job description, “project architect,” and cuts the number of people who hold it.

The line protects the title. Nobody ever paid a mortgage with a title.

A2. The 90-day test.

Diamandis gives executives a test, in his words: “could a frontier model plus a motivated team rebuild your product or service in ninety days?” If the answer is yes, what you have is a head start, not a moat.

He has never run that test on the professional class, so I will. Linda’s compliance work fails it. Her audit work passes, for now: audits require a licensed CPA, and they are required for SEC-reporting companies, many nonprofits and many business sales, so there is real work there. But audits are a small share of all accounting work and nowhere near enough to employ every accountant. It is a moat, and it is a narrow one. A4 picks that up.

In the same piece, he tells the story of a friend whose revenue went “from $160 million to $50 million,” and says of him: “He isn’t depressed; he’s paying attention.” Elsewhere he writes that for anyone who has not yet started a company, “the remaining barrier is almost entirely psychological.”

Put those side by side. The founder who loses two-thirds of his revenue is paying attention. The CPA who is losing hers is told the barrier is in her head. Her barrier is not psychological. It is a P&L.

A3. The missing bottom rung.

Firms are removing the entry level first.

Employment of 22-to-25-year-olds in the most AI-exposed occupations fell about 11% from late 2022 to mid-2026, while the same age group in the least exposed occupations grew about 10% (Stanford Digital Economy Lab, August 2026). Recent-graduate unemployment is about 5.6% and underemployment 42% (New York Fed, second quarter 2026). The Stanford authors call their numbers descriptive, not causal, and so do I. The pattern is still the pattern.

Large law firms expect to grow their associate ranks by 2027, 86% of them, but only 35% plan to expand their first-year classes (Citi Hildebrandt survey, 2025). They want associates. They do not want to train them.

Mike’s firm cut its junior drafters first. He can see that he is next in line, because the work that used to need a drafter under him now needs nobody under him.

A profession that stops hiring at the bottom does not stop. It ages out.

A4. The signature economy.

As AI does the underlying work, fees compress toward the one thing a licensed person still provides: the liability.

The signature becomes the last thing left to sell. A tax return the software prepared in minutes still needs a preparer’s name on it. An audit opinion still needs a CPA’s. What the client is paying for shifts from the work to the willingness to be sued for it. What survives is trust, and Part 5 comes back to that.

An architect’s stamp works the same way. The stamp survives and the hours behind it do not. The firm needs fewer licensed architects to stamp the same volume of work, so Mike’s license does not protect him. It just makes him the last one out instead of the first.

When the signature is all that is left to sell, the price of the signature is the price of the risk, and that is a much smaller number than the price of the work.

B. Why the household cannot absorb it

B1. Sticky income.

The carrying class takes whatever price it is given for its own work.

Raises run about 3 to 3.5% a year, and have for a while: that is the range in the 2025-to-2027 salary-budget surveys from WTW, Mercer and the Conference Board, and the government’s Employment Cost Index was up 3.1 to 3.2% in the year to June 2026. Against that, look at what compounded over the past 25 years: health premiums at 6.0% a year, tuition at 5.0%, home prices at 4.7%, nursing-home care at about 3.3%. Wages ran 3.3%. Overall prices ran 2.6%.

The family health premium went from $6,351 in 2000 to $28,303 in 2026, up 5% last year alone (KFF). From 2000 to 2025 the national average wage went up 2.3 times. Linda pays that premium in full, because she is the employer.

Denver is no exception. Its house price index is up 234% since 2000 (FHFA), close to the national figure of 237% (Case-Shiller).

This answers “they should have saved more.” A 3.3% raise against costs compounding at 5 and 6% is a pay cut with a thank-you card attached. You cannot save a spread that does not exist.

The same holds for Mike. His raise is whatever his firm’s salary budget allows this year, and he takes it.

B2. The wealth cliff.

The cushion starts at the 90th percentile, just above the wall.

Median net worth is $362,500 for households in the 60th to 80th income percentile, $814,600 for the 80th to 90th, and $3,669,900 for the top tenth (Federal Reserve, Survey of Consumer Finances 2025). The top 10% has 4.5 times the cushion of the band just below it, and 10 times the 60th-to-80th. And for the wall, most of that net worth is the house.

Linda is the example. About $820,000 of her net worth is home equity she cannot spend. A critic who calls her a millionaire on paper gets the answer before he finishes the sentence: the money is the house, and she lives in it.

She cannot get at it by downsizing, either. A smaller house at today’s rate costs her about the same per month as the house she has (B5 has the numbers). And selling takes a bite out of the equity before she goes anywhere. Commissions at 5 to 6% of $1.1 million are $55,000 to $66,000. After those costs, her gain over the 2013 price is about $560,000 to $570,000, which is more than the $500,000 a married couple can exclude, so roughly $60,000 to $70,000 of it is taxed: 15% federal, 3.8% net investment income tax, and Colorado’s 4.4%, call it $14,000 to $16,000. That is $70,000 to $80,000 gone before she moves, into a house that costs her the same each month.

The top-tenth median hides a spread, too. The 90th-to-95th band from Part 2 holds plenty of households with a big house, a big mortgage and no cushion at all.

Wealth that cannot be spent is not a cushion. It is a view.

B3. Debt does not deflate.

Cheaper goods do not shrink a mortgage.

Irving Fisher described the mechanism in 1933. When incomes and prices fall, debt that was fixed in dollars gets heavier, because the dollars to pay it are harder to earn. Deflation is a gift to the debt-free and a trap for anyone whose income is the thing getting cheaper.

Linda’s fixed bills are more than the mortgage. The family health premium, about $28,000 a year. The Parent PLUS loan, about $340 a month. Property tax, about $5,800 a year on Denver’s 2025 levy, and homeowner’s insurance, which the state puts at about $3,000 a year on average in Denver County. Her mother’s care. Help for the son whose launch stalled. Not one of those shrinks when her fees do.

Mike bought in 2022 at peak prices, with a mortgage sized for two incomes. Lose one income and the payment stays exactly the same.

Abundance lowers prices. It does not lower balances. The mortgage is not on the list of things that get cheap.

B4. The load she carries.

One Linda holds up more than one household.

More than 51 million Americans, nearly one adult in five, are caring for someone age 50 or older (AARP and the National Alliance for Caregiving, 2025). Fifty-nine percent of parents with children aged 18 to 34 gave them financial help in the past year (Pew, 2024). Twenty-three percent of adults are in what Pew calls the sandwich generation, with a parent 65 or older and a child they still support; among people in their 40s it is more than half. And 58% of 18-to-24-year-olds and 16% of 25-to-34-year-olds live with a parent (Census, 2025).

Linda has both ends of it: her mother’s care and the son whose launch stalled.

None of this money is counted by any official measure. There is no line on a tax return for “carried my mother,” and no survey of consumer finances asks who is paying the rent on the adult child’s apartment. When Linda’s income goes, the official data will record one household losing income. The real count is three.

B5. The exits are closing.

Every way out that worked in the past needed a buyer or an employer on the other side.

Working longer, selling the practice, taking a job at another firm: all three depend on a market, and the valley shrinks the market. Small-firm revenue and hiring expectations fell to their lowest level since the 2020 survey (Federal Reserve Banks, Small Business Credit Survey, March 2026).

Selling the house is an exit too, and it is closing. Linda’s mortgage is at 3%. Today’s 30-year rate is 7.32% (Freddie Mac, October 22, 2026). The median home in the Denver metro sold for $575,000 in September (REcolorado); buy it today with 20% down and the principal and interest come to about $3,160 a month, 2.7 times her current payment. Moving down does not lower the bill. It raises it.

For Mike, the way out is another firm, and every firm is adopting the same tools.

And there is a reason this time is different from every displacement before it. Every past one hit one kind of work at a time, and the people it displaced moved to the kind that was growing. Farm work was 41% of American jobs in 1900 and is under 2% now, and the farmhands went to the factories (USDA, BLS). Manufacturing fell from its June 1979 peak of 19.6 million jobs to 12.7 million in September 2026, and the factory hands went to services (BLS). This one is not confined to a sector. The same tools that do Linda’s compliance work do Mike’s drawings, the paralegal’s research, the radiologist’s first read and the warehouse’s picking. If it plays out as far as the abundance camp says it will, there is no kind of work that AI and robots will not do better than we do. Either way, there is no growing sector standing by to absorb the people who leave, and every exit in this passage assumes one.

The 1994 fear passed because the next sector was hiring. There is no next sector.

C. Why it spreads through the economy

C1. Workers and customers are the same people, and that includes AI’s customers.

The people displaced to create the abundance are the people who were supposed to buy it.

Here is the loop in three people. Linda does the books for the roofer. The roofer puts the roofs on the buildings Mike designs. Mike designed Linda’s office. When the accounting goes to software, the roofer loses a customer. When the drawings go to software, so does Linda. There is no point in that circle where the money comes from somewhere else.

Now apply the same circle to AI itself. Businesses pay for AI out of the payroll it replaces, and their own customers are the people who were on that payroll. AI’s revenue rises in the first round. Its customers’ revenue falls in the second.

There is a precedent everyone already knows, and it points the other way. The first assembly line could not keep its workers. In January 1914, Ford more than doubled their pay, to five dollars for an eight-hour day, up from $2.34 for nine. Annual turnover at the plant had hit 370% in 1913; after the raise it fell to 54%, then 16% (Raff and Summers, 1987). The economists who studied it found the reason was turnover and effort, not sales. But the effect was the mass market anyway. The people on the line became the people in the showroom, and by 1922 Ford was saying so himself, in the chapter of his memoir titled “Wages”: “I believe in the first place that, all other considerations aside, our own sales depend in a measure upon the wages we pay.”

AI has no turnover problem, because it has no workers. The wage that turned mass production into mass consumption is the cost it exists to remove.

Why can’t any one firm stop? Because at the level of a single firm, paying your own workers so they can buy your product never pays. Ford’s raise only worked because turnover made it profitable. The AI version is profitable for no single firm: each one keeps all of its own savings from automating and bears only a sliver of the lost demand, so every firm automates past the point that is good for all of them. Two economists, Brett Hemenway Falk and Gerry Tsoukalas, modeled exactly this in a 2026 preprint called “The AI Layoff Trap,” and the name is the finding. What is rational for each firm is ruinous when every firm does it, and the firm that hesitates loses to the one that does not.

Taken all the way, the abundance destination is a world where work is optional. Elon Musk, November 2025: “My prediction is that work will be optional. It’ll be like playing sports or a video game or something like that.” If that plays out all the way, wage income goes to zero, and every dollar a customer spends has to arrive some other way. The only pool of money large enough is the revenue of the companies that own the AI, and C2 shows their investors are counting on that same money. The destination needs customers who have income and no jobs, and nobody has yet said who pays them.

To be clear about what I am not claiming: I am not saying no one has income. Businesses and the top tenth of households keep buying. The end state is conditional, and I do not predict it. What I am saying is that Diamandis asks who captures the growth (“The question is who captures it”) and never asks who buys it.

Henry Ford’s whole trick was that his workers could afford a Ford. AI’s trick is that it does not need workers. So who is left to afford the product?

C2. The investors need their money back.

Prices falling toward zero and very large investor returns cannot both hold.

AI infrastructure investment is “projected to exceed $1.3 trillion by 2027” across six companies (S&P Global Ratings, August 2026). The four largest cloud companies alone planned about $730 billion for 2026, up about 77% from the year before. The largest pool of money in the economy that could repay spending on that scale is the wage bill: Social Security’s taxable payroll is $11.0 trillion this year (2026 Trustees Report). The investment pays back only if the displacement happens.

So there is a fork. If prices stay near zero, investors lose, and the losses sit in the index funds inside the wall’s own retirement accounts. If the winners consolidate and reprice, the wall’s costs rise. Both branches land on the wall.

Today’s AI prices are being carried by investors. SemiAnalysis, an outside research firm, estimates that flat-rate subscriptions are “just 10% of overall revenue” at Anthropic while they “can take up over 40% of inference compute”; a user who maxes out a $200 plan on the most expensive model runs the company a −369% gross margin, and at a more typical 20% of the limit the plan is about 6% profitable. On the cheaper-to-serve model the same plan is profitable at either usage. These are modeled figures, not company-reported ones, and since this paper was written with Claude, you should know that this passage cites numbers about Claude’s maker. OpenAI, according to shareholder documents reported by two news outlets, took in $5.7 billion in the first quarter of 2026 and lost $9.3 billion in operations.

The first repricing has started. In September 2026, OpenAI reopened its $200 plan with a lower usage allowance and added a $500 tier above it.

Low prices that did not stay low are not hard to find. Netflix’s streaming plan was $7.99 a month in 2011 and the standard plan is $19.99 today. Uber’s accumulated losses reached $32.8 billion by the end of 2022, the year before its first profitable one since going public (its own annual reports), and the price of a ride rose 83 to 92% between early 2018 and 2021-22, by two industry tallies. Two business examples, closer to Linda: QuickBooks Online Plus, the software she and her clients keep their books on, was $70 a month until mid-2021 and is $140 now; the latest increase took effect August 1, 2026, and Intuit’s notice tied it to “new AI-powered capabilities.” And when Broadcom bought VMware for $61 billion and moved customers to bundled subscriptions, an AT&T executive wrote to Broadcom that the proposed increase was 1,050% in one year, an email later filed in a New York court, and European cloud providers reported increases from 800% to 1,500% to the European Commission.

Netflix repriced once it had won on scale. The amount to be earned back here is close to three-quarters of a trillion dollars this year alone, heading past $1.3 trillion in 2027.

Part of the fall in AI prices is real and not a subsidy. MIT researchers find the price of a given level of AI performance has fallen 5 to 10 times a year, about 3 times of that from better algorithms (Gundlach, Lynch, Mertens and Thompson, “The Price of Progress,” preprint, 2025, revised 2026). So I am not saying every token is sold below cost. I am saying the companies are being carried by their investors, and investors do not carry anyone forever.

C3. The cascade.

The damage moves in six stages: lost income, lost customers, lost tax revenue, lost family support, lost housing and credit, and finally lost asset prices. Each stage feeds the next, and the whole thing moves slowly enough that no stage looks like an emergency.

That is the boiling-frog risk. A slow grind never trips an emergency threshold. I present this cascade as a hypothesis, and the timeline is its weakest part. The argument holds even if the valley is shallow.

Here is where the grind shows up first. When income wobbles, a household cuts in a fixed order: the mortgage, the groceries, the kids and the parents are kept, and everything discretionary goes. The cheapest things go earliest, because they are the easiest to drop. So the valley is visible in book sales, restaurant checks, gym memberships and youth sports registrations long before it is visible in unemployment, and the official instruments in Part 4 are watching the wrong end of the household budget.

Tim Ferriss published his own numbers in June 2026: print sales of his five-book catalog fell 46% in 2025 and were running 57% further down in 2026 year to date. Self-help as a category fell 26.3% in the first quarter of 2026, against 8.8% for all adult nonfiction (Publishers Weekly). He attributes his drop to AI answering the questions his books answered, and I do not dispute that. I also do not claim his numbers show the valley. The point is that from the outside the two causes look exactly the same. A $30 how-to book stops selling whether the reader got the answer from a chatbot or stopped having $30 to spare. A category that is both substitutable by AI and discretionary gets hit from both sides at once, and nobody can tell which side is which until the mortgage data arrives, years later.

This answers “if this were happening, we would see it.” We would see it, in the categories no economist tracks as a signal. Part 6 asks someone to start tracking them.

The valley shows up in the Little League registrations before it shows up in the unemployment rate.

C4. The payroll spiral.

Social Security’s rescue will arrive late, from a smaller payroll, at a higher rate on the people still working.

The Congressional Budget Office puts the retirement trust fund’s exhaustion in fiscal 2032, and the 2026 Trustees Report agrees, with 78% of benefits payable after that. Both projections sit on a baseline that assumes taxable payroll keeps growing. A software subscription that replaces a bookkeeper pays no payroll tax.

Run a 2008-sized hit to payroll through CBO’s 2032 numbers, holding benefits where they are, and the gap the tax has to close goes from 3.5 points to about 4.7, a third more per worker. That is my arithmetic on CBO’s figures, not a CBO projection, and I want to be careful about that.

Congress has never let a Social Security check bounce, and it will not start. The rescue is the base case. The question is what the rescue looks like when it comes: late, from a smaller base, at a higher rate, on the carrying class, who cannot pass a tax increase through to anyone. Linda, as an S-corp owner, pays both halves. The subscription that replaced her bookkeeper pays neither.

C5. The state and local spiral.

Washington can be late. States cannot. They have to balance their budgets in the same year the tax base shrinks.

After 2008, 36 state unemployment programs ran their funds dry and borrowed from the Treasury to pay claims, with $51.2 billion outstanding at the peak in May 2011. Colorado lost about 154,000 jobs (6.5%) from May 2008 to January 2010, and the country lost 8.7 million (6.3%). State and local governments then shed about 747,000 jobs of their own over the next five years (BLS). Colorado’s general fund is about 58% individual income tax and 27% sales and use tax (FY 2024-25 actuals, Legislative Council Staff). Take the income and the spending out of the carrying class, and you have taken out most of the state.

Some taxes leave with the job itself. Denver’s occupational privilege tax is $5.75 a month from the employee and $4 from the employer, $117 a year per job. Greenwood Village ($2 and $2), Glendale ($5 and $5) and Sheridan ($3 and $3) levy their own, and Aurora repealed its tax as of January 1, 2025. Colorado’s paid family and medical leave premium is 0.88% of wages in 2026, falling to 0.86% in 2027, split between employer and employee at firms with ten or more workers; a firm of Linda’s size remits only the employee half. A software subscription pays neither.

Colorado does not tax hosted software today. A bill signed in June 2026 repeals the exemption for software “available for repeated sale and license” starting January 1, 2027, so some of it becomes taxable at the 2.9% state rate; whether a cloud subscription falls inside that definition is not yet settled, because the Department of Revenue has not issued its rule. Even if it does, that is a few percent of a small subscription price, set against the wage-based taxes the job it replaced used to pay. A subscription is a rounding error next to a salary, so even a steep tax on it collects a rounding error. The state reaching for that new base is itself part of the spiral.

I watched the last one from the inside. Across about 300 payrolls at Easytrack, the employers who kept their people on through 2009 and 2010 paid the unemployment-insurance surcharge for the ones who laid off. The system billed the survivors for the casualties.

C6. The payroll-less recovery.

Every past crash created new businesses that hired. This one creates businesses designed not to.

Only 6.9% of new business applications from January to August 2026 said they planned to pay wages, and the share fell every month, from 7.4% in January to 6.4% in August (Census Business Formation Statistics). Of the business establishments opened in 1994, only one in five was still open 20 years later (BLS).

The small firm’s one real advantage was speed, and AI hands it to the large incumbent. A 4,000-person company can now move like a four-person one, and it still has the balance sheet, the brand and the customer list. And the founder who failed once does not do better the next time just because the tools got better.

I started Easytrack in 2007 and hired into the slack labor market of 2009. That is what recoveries used to look like: somebody started something and needed people. The recovery Diamandis describes is five times as many companies on a fifth of the staff, and 93 of every 100 of them not planning to hire anyone at all.

This closes the contradiction in Part 1. Five times more companies at zero payroll is zero payroll.

D. Why reaching abundance does not fix it

D1. Abundance is a price. The wall’s problem is a balance sheet.

Every item on the abundance list is a price falling toward zero. Energy, food, transportation, information, eventually housing. And if housing does get cheap, it takes down the value of the one real asset most households own, which is the home. The mortgage is not on the list.

On Moonshots #301, Dave Blundin made the same point from the other side: abundance will show up as falling prices, not as rising GDP. He is right, and that is exactly the trouble. A falling price helps with what you buy. It does nothing for what you owe.

Abundance shows up in what a household spends. The wall’s trouble is in what it owes and what it owns. Cheaper groceries help Linda. They do not touch the $280,000, the $30,000, the premium or the tax bill, and they do not replace the $235,000 that paid all of it.

You can give a family every price cut in the world and still leave it with the same debt and no income. That is not abundance. That is a cheaper place to be broke.

D2. Income first, costs last, debt never.

Abundance reaches the wall’s income before it reaches the wall’s costs.

Information work gets cheap first, because that is where the models already are, and information work is the wall’s revenue. Health care, tuition, elder care and housing get cheap last, because institutions set those prices, not technology. The gap widens inside the valley.

The record so far says the same thing. Since 2000 the family health premium is up about 345%, in-state public tuition about 240%, home prices about 237%, and a private nursing-home room has roughly doubled since 2004. Wages are up 126%. Overall prices are up 87%.

Why these prices are set by institutions: Medicare’s funding for medical residencies has been capped at 1996 levels since 1997, with 1,200 slots added since 2021, which caps how many doctors get trained. “Pay-for-delay” deals keep generics off the market about 17 months longer and cost consumers about $3.5 billion a year (FTC). Insulin is the household example; Lilly cut Humalog’s list price 70% in 2023 after years of increases. The first gene-therapy cures arrived at list prices of $2.1 to $3.1 million each (Zolgensma in 2019, Casgevy and Lyfgenia in 2023, per the makers’ own filings). Energy is the same pattern: the cost of electricity from large solar farms fell about 90% from 2010 to 2024 (IRENA), and the residential electricity price rose 5% in 2025 anyway. I am not claiming cures are being suppressed. I am saying the cure comes priced for the top, and the price comes down on an institution’s schedule, not a chip’s.

Some costs rise first. Building the AI is bidding up the inputs before AI lowers the price of anything Linda buys. The Energy Information Administration expects the average residential electricity price to reach 18.2 cents a kilowatt-hour in 2026, up 5.3%, and 18.7 cents in 2027. On PJM, the largest U.S. grid, the capacity price has cleared at its federal cap three auctions running; in one of them, nearly 5,100 of the 5,250 megawatts of added forecast load were data centers, and the latest auction came up 6,831 megawatts short of the reliability requirement, the first grid-wide shortfall. DRAM memory chips: contract prices rose 93 to 98% in the first quarter of 2026 and were forecast up another 58 to 63% in the second, as chipmakers moved capacity to AI (TrendForce). Supply stays tight through 2027. A large gas turbine ordered today is about three years out, and the biggest maker has sold most of its slots through 2030 and more than half of 2031.

These are small lines in Linda’s budget, and I say so. U.S. phone prices are up about 5% this year against 21% in India, so the heaviest burden falls outside the wall, and how much of the rise in electric bills comes from data centers is contested. I am not saying AI is inflationary. I am saying the price of what AI produces is falling while the prices of what it consumes are rising, and the wall meets them in that order.

D3. Who gets what.

The bottom half gets a real improvement: a floor of basic needs at a falling price, which is the part of the abundance case I believe most. The top tenth gets the revenue, because it owns the machines. The wall gets hit from three sides: its income goes first, its costs fall last, and its debt never moves.

The realistic outcome for most of the wall is a cheaper, smaller life with the house kept. That is what Linda will get, and Mike, if he is lucky. It is not a catastrophe. It is a demotion, for the people who did everything they were told to do.

Which brings back the thing the floor leaves out. The middle-class promise was never “your needs will be met.” It was “you can climb.” A floor does not answer that. The only answers I have found to it are the ones where people own a piece of the machines, and those are rows 9 and 10 in the table in Part 5. I do not argue for them there. I only note that they are the only rows that answer the question.

Part 4: Why it has only been half-named

Diamandis named the years in March. What he did not name is who is in them and what they owe. The reasons are listable, and listing them matters, because otherwise a reader takes the silence as proof that nothing is wrong.

1. The flaw of position. The people writing the abundance message have already crossed the valley. His September 2 piece is addressed to employees, entrepreneurs, parents, investors and policymakers. His September 24 piece speaks to the entrepreneur, executive, investor, student and parent. Neither list, and not his March valley either, has a place for a 52-year-old with a four-person firm. That is a fact about where they stand. It is not a moral failing, and I am not treating it as one.

2. Net numbers hide it by construction. A net gain of 78 million jobs says nothing about whether the new jobs arrive in the same year, the same city or the same age bracket as the lost ones. In the WEF report he draws on, the largest gain is farmworkers, 35 million of them, with delivery drivers and personal care aides close behind. His piece names no occupations at all. A net number is a true number that answers a question nobody in the valley is asking.

3. The instruments cannot see it. The safety net is built to see a layoff. The valley does not produce one.

4. The cost has no address. A data center’s cost lands on a few hundred neighbors who share a county meeting, so they organize. The wall’s cost arrives one lost client at a time across 36 million small businesses, and there is no meeting for that. On Moonshots #301, Diamandis said that aviation rules get written after the accidents. A plane crash has a wreckage site and an investigation. A CPA who loses her clients one at a time has neither, so nobody writes the rule.

5. It does not fit the script. Displacement is supposed to happen to factory workers. Linda is educated, well paid and good with the tools, so both political coalitions assume she is secure. I will keep this structural and leave the parties out of it, but the assumption is bipartisan and it is wrong.

6. The people in it have no time to write. Linda is not the letter-writing type, and most of the wall is too busy carrying to speak. The people with the time to write about the transition are, almost by definition, the people who have already crossed it. I am writing, and I almost didn’t, because of the opportunity cost. It is time-consuming, and it is not easy to write well. I had used Claude for enough other things to be convinced it could give me the leverage to finish a project like this, and to finish it well. That is the only reason this paper exists.

7. It is nobody’s field. There is a large literature on AI capability, on AI safety and on the aggregate economy. There is almost nothing on how a household, a four-person firm or a mid-career professional gets through the next eight years. On the same Moonshots episode, Emad Mostaque said that most of what people fear from AI is already covered by some regulation, and that there is no map of the rest. Linda’s problem is in the unmapped part. This paper is my attempt to draw that piece of the map.

8. What Diamandis has already conceded. He has written, citing Goldman Sachs, that professional services firms, “consulting, law, accounting,” face existential risk. He has written that AI “can’t negotiate, build trust, make judgment calls under uncertainty, or read a room.” He has written that entry-level roles like data entry and junior paralegal work “are being vaporized.” He has put a five-year window on the founders who get hurt. And in March he wrote that there is a valley, 2026 to 2031, that “must be named.” The naming is more than half done, by the other side. What he has not done is put the carrying class in the valley or the debt on the bridge.

9. Why naming it matters now. Whoever names the problem writes the story. A backlash has already formed around data centers and electricity bills, and it contains no Lindas; it is about the neighbors with the county meeting. If the abundance side does not write the transition story, a darker and less true one gets written, by people who are against the whole project. I would rather the story be written by people who want abundance to work.

Part 5: Possible solutions, a starting list

A good diagnosis is half the cure. This paper is the diagnosis, and I am asking for help with the other half.

This part is a list of twelve possible solutions, each with what it is, where it helps and where it falls down. It is here to get people thinking and to be tested, built on or replaced. No row is recommended. Some of these may not work. Each row also explains the problem one more time: where a solution falls down is one more side of the valley, the mortgage it cannot reach, the trigger it cannot see, the vote it cannot wait for.

The test any cure has to pass. Four questions, so that you can judge these rows and your own ideas the same way:

  1. Does it reach the debt, and not only the income?
  2. Does it keep money circulating, and not only set a floor?
  3. Can it start without waiting for Washington?
  4. Does it work through the carrying class, and not around it?
Possible solution What it is Where it helps Where it falls down
1. Conversion A firm like Linda’s stops selling compliance work and sells guidance through the transition to the 150 clients who already trust it. A worker like Mike sells his expertise directly, for example as an owner’s representative. It can start now and needs no one’s permission. It works through the carrying class. Financial advisors have made a similar move as trading commissions went to zero; 72% of advisor pay is now fee-based. It works for the most adaptive firms only, and demand for this advice is unproven. It does not touch the debt.
2. Household moves made from strength A household locks in or extends its loan terms and turns some home equity into a cash reserve while its income and credit are still good. It is available today, one household at a time, and it reaches the debt. It only works before trouble shows. It is costly at today’s rates, and it buys time without replacing income.
3. Circulation A mutual-credit network that runs alongside the dollar economy, where members trade work for work. AI does the matching and CPAs keep the ledger. It keeps local trade moving when dollars are short, and it uses skills the carrying class already has. Switzerland’s WIR Bank has run one since 1934. It does not pay a mortgage, a tax bill or an insurance premium. Barter income is taxable (IRS Topic 420), so it needs a carve-out.
4. Local purchasing Hospitals, universities and governments redirect their buying to local firms. It keeps money circulating locally without federal action. Preston, England is the model. It only works where large institutions exist, and it reaches only the firms that sell to them. It does not touch the debt.
5. Mortgages that adjust (new loans) An opt-in mortgage, priced up front, whose payment adjusts when a named trigger is hit. Government reinsures only the catastrophic layer. It reaches the debt. Nothing is changed after the fact, so lenders can price it. It only helps loans written from here on. Linda will not trade a 3% mortgage for one at 7.4%. It needs the mortgage agencies and is slow to build.
6. A refinancing agency authorized in advance (existing loans) Congress writes the rules and the triggers now. When the triggers are hit, the agency buys distressed mortgages and refinances them on longer terms, and forgiven debt is exempt from income tax. It reaches the mortgages that already exist, including Linda’s and Mike’s. The Home Owners’ Loan Corporation of 1933 refinanced just over a million mortgages and closed its books in 1951 with a small surplus. It needs Congress, so it fails the third test. Its triggers must be ones a slow grind actually moves.
7. State work-sharing A state pays partial unemployment benefits when an employer cuts hours across the board instead of cutting jobs. It keeps people on payroll and keeps them customers. It runs on a system states already have: 31 states, Colorado among them, have a program on the books. Germany’s version carried 6.8 million workers in April 2020. It is built for a temporary slump, not for work that is gone for good. It helps Mike more than Linda, whose loss is a lost client. State funds ran short last time.
8. A compute-layer tax with a dividend A tax on the computing that replaces payroll, paid out to households. It is the one tax base that grows during the valley, and it answers the question of who pays the customers. Alaska’s dividend ($1,200 in 2026) is the model. An academic model of the automation race (Falk and Tsoukalas, 2026) finds that, of the fixes it tests, only a tax on automation works. It is the furthest from passage. It sets a floor and does not reach the mortgage.
9. A public or broad stake in the compute layer People hold ownership shares in the AI infrastructure, so lost labor income is replaced by capital income. It answers the mobility question, because owners can still climb. Nobody has designed how the shares are issued or paid for. It needs Congress or the companies to agree, and the income arrives slowly.
10. Fees to equity A standard agreement that lets a small firm take part of its fee as an ownership share in the clients it carries through. It turns trust into ownership, and firms can start now. It does not pay this month’s bills. The shares are hard to sell and sit in the same local economy. CPA independence rules bar it for audit clients.
11. The Carrying Load Index An open measure of what a household carries in debt, family support and dependents. It makes the load visible to policy, and anyone can start building it now. It measures the load and relieves none of it.
12. Augmentation over replacement Tax and procurement rules that make keeping a professional on payroll with AI tools cheaper than replacing the professional with the tools: a credit or payroll-tax offset for AI spending tied to retained headcount, and public buyers that score bids on it. It works through the carrying class and keeps people customers. It attacks the specific price gap that C5 names, a subscription that pays no payroll tax against a salary that does, and that the 1994 paper named in an earlier form. Firms can adopt the posture now; the rule only widens it. It needs Congress or a state, so it fails the third test. Headcount rules invite gaming. It slows the hit; it does not touch the debt, and in a race the firm that keeps the person loses to the one that does not (C1).

The three debt rows work together. Rows 2, 5 and 6 each cover what the others miss. A household can act now, while its income and credit are still good. A new kind of mortgage covers the loans written from here on. An agency authorized in advance covers the loans that already exist, which is where Linda and Mike are. I say that in three sentences and I do not argue it.

Rows 7, 8 and 9 are also the three candidate answers to the question in C1: who is the Ford of AI? Which firm, or which rule, closes the loop between the people AI displaces and the people AI needs as customers. None of the three exists yet.

Two that look like answers and are not.

  1. Changing loan terms after the fact. It was tried in 2009. The program was meant for three to four million homeowners and started 1.7 million permanent modifications. Lenders answer a surprise change by raising the price of every future loan.
  2. Bankruptcy. It destroys credit, it can threaten a professional license, and a Chapter 13 court cannot modify the mortgage on a primary residence.

A new idea becomes a row here or a question in Part 6. That is the whole editorial policy.

Part 6: Open questions and other ideas

These are the questions this paper names and hands to others. Each gets a paragraph at most and no further development in version 1.0.

Measure it

  1. The right trigger. A set of indicators a slow grind actually moves, such as labor-force participation, graduate underemployment, small-firm revenue expectations and loan delinquency. Also worth testing: discretionary categories with a professional-class customer base, such as nonfiction book sales, restaurant checks and youth sports registrations, since the household cuts those before any official measure moves (see C3).
  2. The sticky-income chart. Twenty-five years of wage growth against the cost of health care, tuition, elder care and housing, on one page. The data are all in B1 and D2. Someone should draw it.

Prove it or break it

  1. The demand side, or: who is the Ford of AI? Which firm, or which rule, closes the loop between the people AI displaces and the people AI needs as customers. Who buys anything during the valley, with numbers, and where customers’ income comes from in the end state if work is optional. Ford’s answer in 1914 worked for one firm because turnover made it pay; the AI version pays for no single firm. Rows 7, 8 and 9 in Part 5 are three candidate answers, and none exists yet. This is the paper’s most exposed point and the first place I want help.
  2. Whether trust survives when AI agents choose the accountant. Part 3 says what survives is trust. If a client’s AI agent picks the CPA, whose trust is it?
  3. The size of the state and local spiral, starting with Colorado.
  4. The timeline. Which signals would show the valley is shallower or deeper than argued here.
  5. The return AI investors need, sized against the national wage bill.
  6. The buildout’s own price effects. How much of the rise in memory, electricity and power-equipment prices traces to AI demand, who pays it, and when the new supply arrives.
  7. Whether any public priority on AI’s uses would reach the household, or only re-order which professions are displaced first.

House it

  1. A permanent home for the work: a numbered archive, a forum and later a wiki.

Held for later versions. His own “valley-bridging mechanisms,” the $3,000 replacement income and the automation dividend, costed against the carrying class’s load, which is a paper in itself. Minimax regret for a household that cannot diversify one career and one mortgage. Whether any AI lab or hyperscaler is large enough that its customers’ income becomes its own problem, which would make it the candidate Ford. The eight-hour-day half of the 1914 story. Whether a slower frontier, the pace the safety camp argues for, shortens the valley or lengthens it. The regulation questions belong to a different paper and stay out of this one.

Part 7: The handoff

What this version is. Version 1.0, dated, with gaps left on purpose. I wrote it to start something, not to run it. I have other work that needs me, so I am not promising later versions or maintaining a forum. The paper is licensed so that anyone can take it from here, and I hope someone better placed than I am does. If you build on it, put your name on your work and a link to this one. I am on LinkedIn.

The model. The early internet was built on open requests for comment. Somebody wrote down a problem and a first design, numbered it, and asked everyone else to make it better. This paper is one of those. Here is the problem, and here is a list of places to start.

What to do with it. Take it. The license at the end of this part lets you copy it, quote it, build on it, argue with it or rewrite it, with credit. Write your own version for your profession or your town. Run one number through your own books. Cost one row of the table in Part 5. Take one question in Part 6 and answer it. Find the hole in Part 3 and publish the correction. Hand it to the person who could have written it better than I did. Put your name on your work and a link to this one, and post it where your people are, not where mine are. Nothing in this paper needs my permission or my reply, and the list above is a starting point, not a boundary. Anything that moves the question forward counts.

What I am asking of the abundance community. Three things, and none of them is to stop. Put voices from inside the transition on the show. Give the transition a track at the summit. Add the journey to the core message, next to the destination. You have named the valley. Help name the people in it.

Why with superintelligence. The technology causing the valley is also the best tool for bridging it. This paper was written that way, with the same tools that are doing Linda’s compliance work, and I think that is the only honest way to write it.

A note from 1994. In March of that year I wrote an MBA paper on job security in the downsizing era. It argued that U.S. policy made layoffs cheaper for employers than their true cost, and it recommended short-time compensation in every state, stronger experience rating of unemployment taxes, and health coverage after a layoff. Three of those are rows or mechanisms in this paper. The problem was named then too, for a different machine. The 1994 paper is posted beside this one, for anyone who wants to see how little the policy toolkit has moved in thirty-two years.

License. This paper is published under a Creative Commons Attribution 4.0 license (CC BY 4.0), so that anyone can share it, build on it or adapt it, with credit. Please do.

Sources. Every number in this paper is sourced. The full list, with links, is posted at thevalleypaper.com/sources, numbered in the order the figures appear. Where a figure rests on a secondary report because the primary could not be opened, the list says so.

We cannot direct the wind, but we can adjust our sails.

Brian Senjem
Pine, Colorado
October 28, 2026

© 2026 Brian Senjem. Published under a Creative Commons Attribution 4.0 license (see Part 7). Sources: /sources. The 1994 paper referenced in Part 7: /1994.