The AI Checkpoint: What the Data Says About Work, Wealth, and Companionship in 2026

A data-first snapshot of how Humans and AI are actually living together.

Four months ago, The Great De-Coupling closed with a promise: follow the data wherever it leads.

This is that promise taking shape. Welcome to the Checkpoint, a twice-a-year, data-first snapshot of how humans and AI are actually living together. The work we do. The wealth we build. The company we keep. The strange, the hopeful, the hard to believe. And when fresh data starts answering something I wrote earlier, it goes on the scoreboard, whichever way it points.

One number anchors this first checkpoint, and if you remember nothing else, remember this one. For as long as the modern record has been kept, a college degree meant you were less likely to be unemployed than the average worker. In 2026, that flipped. Recent graduates: 5.6 percent unemployed. Everyone else: 4.2. I call it the Inverted Ladder, and it is the single clearest measurement we have of what AI is doing to work.

Here is what the data says in August 2026.

Work: The Break Started at the Bottom Rung

If you read only headlines, AI has gutted the workforce. If you read only aggregate statistics, AI has barely touched it. Both are wrong, and the truth is more interesting.

The aggregate numbers look calm. Overall US layoffs are declining in 2026. A National Bureau of Economic Research paper found 90 percent of executives report zero AI impact on employment at their companies. If you stopped there, you would call the whole thing hype.

Do not stop there.

At the entry level, where labor is most substitutable, the data is anything but calm. Salaried postings requiring no prior experience have fallen roughly 73 percent in four years. And then there is the Inverted Ladder: recent college graduates now face higher unemployment than the general workforce, 5.6 percent against 4.2 percent. For decades the degree premium meant graduates fared better than everyone else. That deal is showing initial signs of change. Employers even have a new word for what they are doing to junior roles: seniorization, loading entry-level postings with senior-level judgment requirements, because AI absorbed the routine work juniors used to learn on.

The ladder did not disappear for those already on it. It disappeared for those about to step on.

And the layoff data itself? Polluted, in both directions. Fifty-six percent of 2026 layoff announcements cite AI, yet 59 percent of hiring managers privately admit AI was cover for overhiring corrections and cost pressure. Even Sam Altman conceded that almost every company doing layoffs is blaming AI, whether or not it is about AI. Meanwhile, real AI-driven cuts hide inside vague "restructuring" language. Anyone claiming they know exactly how many jobs AI has eliminated is selling something.

Money: The Fork Between Enterprises

While everyone watched the line between humans and machines, the sharpest divide of 2026 opened somewhere else entirely: between enterprises.

BCG finds 60 percent of companies investing in AI are generating no material value from it. Five percent are creating substantial value at scale. That is not a normal spread of outcomes. That is a fork in the road, and the fork is not about who bought better AI.

McKinsey tested dozens of factors to explain who actually captures value. The strongest single contributor was not model choice, not spend, not talent density. It was intentional redesign of workflows. Companies that rethink how work flows end to end are 24 percentage points more likely to see measurable impact. High performers are three times more likely to redesign than to deploy tools into existing processes. Seventy percent of failures trace to people and process, not technology.

I see this every week in my work with large enterprises, and I have watched it play out in a small business I know intimately. The pattern is public now, and it is stark.

Teams that bolt AI onto existing processes are churning friction and waiting for value that never arrives. Teams that step back and reimagine the workflow end to end are compounding multiple-x gains.

We have seen this movie before. In 1990, Michael Hammer wrote in Harvard Business Review: "Don't automate, obliterate." Companies that paved their cow paths with software got faster cow paths. Ford reengineered accounts payable from 400 clerks to five, not by automating invoice matching, but by redesigning the process so most matching never needed to happen. Hammer's insight waited thirty-six years for a technology powerful enough to deserve it. Bolt-on AI is the new paving of cow paths.

So here is the test I use, and I am giving it a name: the Compounding Inequality. An AI initiative deserves funding only when it satisfies one condition:

Time + Energy + Money invested < Short-term wins + Medium-term gains + Long-term compounding

The left side is what the initiative consumes: people's hours, organizational attention, and budget. The right side is what it returns, across three horizons. And here is the part that decides everything: the right side only compounds under one configuration, when the workflow is redesigned so agents own the routine and humans move up to higher-value work. Bolt-on AI pays the left side in full and collects only the first term on the right, a few short-term wins that flatten within a quarter. A redesigned workflow keeps collecting, because freed human capacity creates the next improvement, which frees more capacity, year after year.

The exercise for this checkpoint: pick your five most expensive workflows, trigger to outcome, the whole chain. For each one, estimate six things: volume of cycles per year, FTE-hours per cycle, the count of handoffs and rework loops, the hours of senior talent trapped in routine steps, the fraction of steps that genuinely require human judgment, and whether the data feeding it can be trusted. Then run the Compounding Inequality on each, and ask one question: are we paving this cow path, or rebuilding the road? If your AI portfolio is a list of tools attached to unchanged processes, you now know which side of the 60/5 divide you are on. I am developing the full method, Workflow Arbitrage, in a coming post. The name is literal: the gap between what a workflow costs you and what a reimagined version would cost is a mispricing, and it closes when your competitor finds it first.

Love: The Frontier Nobody Budgeted For

Now the part of the snapshot no earnings call covers. While enterprises argued about workflows, ordinary people quietly moved AI into their inner lives.

Common Sense Media's national survey found 72 percent of American teens have tried an AI companion, and half use one regularly. One in three has taken something serious to the AI instead of a human. This is not a niche. This is a generation forming attachment habits on synthetic empathy.

The research is genuinely split, which is exactly why it belongs in a data-first snapshot. Harvard Business School found AI companions reduce momentary loneliness about as well as talking to a person. A 12-month study of 2,000 people found the opposite over time: sustained companion use predicted increased loneliness, likely because it displaces the harder, richer work of human connection. Both findings can be true. A painkiller is not nutrition.

AI companionship works like a painkiller. Real relief, measurable, immediate. And nobody ever got healthy on painkillers alone.

I will not moralize here; the data is young and so is the phenomenon. But a snapshot of humanity with AI that only counted jobs and dollars would be dishonest about what is actually happening in bedrooms and group chats around the world. We will keep watching this line closely.

The Story That Made This Personal

In 2020, my mother visited Seattle for the first time. On her second day, severe abdominal pain took us to the emergency room, where imaging found a kidney stone. It passed with routine medication, and we thought that was the end of it.

It was the beginning. For seven years, she has lived in two separate medical storylines. In one, a rheumatologist manages her joint pain, calls it early-onset arthritis, expected at her age, prescribes pain management, and tells us there is no cure. In the other, a urologist treats kidney stones that recur every six to twelve months and notes that her blood calcium runs high. Two conditions. Two specialists. Two treatment plans. Nobody connecting them. I kept asking the question any son would ask: what is the root cause? The honest answer from both rooms, for seven years, was a shrug. She has eaten the same vegetarian diet her whole life. Calcium does not simply rise in your blood in your sixties for no reason.

Recently, I gave her 2024 lab reports, the same reports her doctors had reviewed, to two AI models. Both, independently, pointed to the same thing: primary hyperparathyroidism. One of four tiny glands in the neck, quietly overproducing the hormone that pulls calcium out of bones and into the bloodstream. Weakening bones in one storyline. Recurring stones in the other. One cause for both, hiding in plain sight between two specialties for seven years. The models also laid out the confirmatory tests to run and what treatment looks like. A physician has since concurred with the diagnosis, and she is now scheduled for a fluorocholine PET scan to localize which gland is overactive. The likely fix, after seven years of managed symptoms, is a routine outpatient surgery.

I do not share this to diminish her doctors. Each was skilled inside their own lane. That is precisely the point. The silos are the disease. Human expertise is deep and narrow by design, and our institutions, hospitals, governments, enterprises, are built from those narrow lanes. An intelligence that reads across every lane at once does something no single specialist can: it connects the dots.

And it is not just medicine. In the De-Coupling post I told you King County took two months to assign an inspector to my septic application; it eventually passed in three and a half. I am now in the building-permit phase, and the clock has not even started: a week to schedule the geotech visit, two weeks for the report, weeks of back-and-forth to get the retaining wall data to the civil engineer, a month or two for reviews. A year in permits, and mine is the average case. The bottleneck is not the people. It is the structure they work inside, the same fragmentation my mother lived in for seven years, wearing a hard hat instead of a hospital gown.

The silos are the disease. AI is the first technology that can read every lane at once, and that is the real promise: better health, better services, and our creative energy back.

Hold the teenager with the AI companion and my mother's diagnosis in your mind at the same time. That is what living with AI actually looks like in 2026, and it is why this snapshot will never be a single storyline.

The Scoreboard: Revisiting Earlier Calls

The format is fixed, the entries are not. Each checkpoint scores whichever earlier claims the data has started to answer, in three parts: what I claimed, what the data says, and one verdict from a set of four. Holding. Mixed. Missed. Too early. No other grades exist.

From The Great De-Coupling: the break shows first where labor is most replaceable. That is precisely where it is showing. The Inverted Ladder and the 73 percent collapse in no-experience postings are early structural evidence, and they arrived faster than I expected. The aggregate economy, meanwhile, remains calm: no demand shock, no macro break. The larger De-Coupling story is a long-horizon one; what this board tracks are its early signals. Verdict: holding at the entry rung, too early everywhere else.

From Signal vs. Slop: the flood arrived, and so did the filters. When I wrote about agency in the synthetic age, the synthetic age was a forecast. It is now a measurement. Graphite's analysis of 65,000 published articles found AI-generated pieces crossed 50 percent of new web content, up from about 10 percent before ChatGPT. But the same research carries a twist worth celebrating: most of that slop never surfaces in Google or ChatGPT results. The flood is real, and the filters are holding better than feared. Curation, taste, and agency did become the scarce skills. Verdict: holding.

The enterprise fork. Nothing I wrote anticipated that the most consequential divide of 2026 would run between companies that reimagine work and companies that automate inertia. It is now the strongest signal in the entire dataset, and it earned its own section above. Verdict: missed, and gladly claimed now.

Where the humans go. As agents absorb routine execution, enterprises are repricing exactly what you would expect: routine down, judgment up. The seniorization data is brutal for those locked off the ladder, and it is also the clearest evidence that the market now pays for what only humans do well. Whether we build the bridges from one side to the other is the question of the decade. It deserves its own post, and it will get one. Verdict: too early, watching closely.

Until the Next Checkpoint

The three scenarios post is coming next: managed transition, concentration and conflict, or post-scarcity leapfrog, with honest probability estimates. The comment thread on the De-Coupling post is still open, and your reasoning still sharpens mine.

This checkpoint will return in roughly six months, same rules. Data first. Predictions scored only when the data speaks. No victory laps.

Until then, one question, and I genuinely want your answers in the comments:

In your own life, where has AI actually changed things this year, your work, your team, your family, maybe even your relationships? Not the headlines. Your life.

That is the dataset nobody publishes. Let's build it here.

Sources: Common Sense Media, "Talk, Trust, and Trade-Offs" (2025). De Freitas et al., "AI Companions Reduce Loneliness," Harvard Business School. BCG, "Companies Must Go Beyond AI Adoption" (2025). BCG, "Scaling AI Requires New Processes, Not Just New Tools" (2026). McKinsey, "The State of AI" (2025). Graphite, "More Articles Are Now Created by AI Than Humans" (2026). Forbes (Aug 2026) and Washington Monthly (May 2026) on entry-level jobs. The Interview Guys on 2026 layoff attribution. Michael Hammer, "Reengineering Work: Don't Automate, Obliterate," Harvard Business Review (1990).