On July 6, 2026, roughly 1,600 people at Xbox opened an email titled "Resetting XBOX." It announced the most significant restructure in the division's history: about 3,200 roles gone through the next fiscal year, a fifth of the entire division, four studios spun out to new owners. Across Microsoft the same day, 4,800 jobs were eliminated, on top of roughly 15,000 the year before. The memo promised a flatter organization built around "player-coaches" and "directly responsible individuals," a cleaner code base, and 50 percent less vendor spend.
Microsoft's HR chief was careful with her words: the departing roles "are not being replaced by AI." Then the sentence that did the real work: "what is true is that AI is changing how work gets done."
Both halves are true, and the second half is the one that matters. Nobody installed a chatbot in an ex-employee's chair. What happened is quieter and more structural. The company committed roughly $190 billion to AI infrastructure this year, its investors punished the stock anyway, and the arithmetic of a smaller headcount producing the same output with AI tooling became the plan rather than the fear. Microsoft was not alone. Around 154,000 tech workers lost their jobs in the first half of 2026, at Meta, Oracle, Amazon, and a long list of companies making the same trade.
For the people who stayed, the transition produced one of the strangest artifacts of the era. Inside Meta, employees burned through roughly 74 trillion AI tokens in a single month, about $221 million worth, while an internal dashboard nicknamed "Claudeconomics" ranked the top 250 consumers. People competed for titles like "Token Legend" and "Cache Wizard." Some reportedly left agents running unattended for hours, purely to climb the board. Shopify, which had run the first leaderboard of this kind, quietly renamed it a usage dashboard and wired in circuit breakers to catch runaway agents, its head of engineering conceding that ranking people by tokens invited competing rather than shipping.
Measuring tokens burned is measuring lines of code with extra steps. The industry spent thirty years learning that inputs are not outcomes, then rebuilt the same vanity metric the moment the input changed. A token consumed is not value created. It is cost incurred. The leaderboards died within months, as they deserved to, and the companies that ran them quietly pivoted to asking the only question that was ever worth asking: what did the work produce?
That question is the hinge of this whole era, and it is exactly the hinge Philipp Hölke identified in Software is for Machines. Humans want Outcomes. His argument is about software vendors: SaaS sold access and called it service, and when the buyer becomes an agent, margin flows to whoever actually delivers the outcome. But the same inversion applies to careers, because a salary is a per-seat license. An employer buys access to your hours and stitches them into outcomes somewhere above your head. When execution gets cheap, the number of seats required to produce the same outcome falls, and no amount of loyalty, tenure, or token burning changes that arithmetic.
Now read the graph from the other end.
While the layoff announcements stacked up, the US Census counted 30.4 million nonemployer firms, companies of exactly one person, generating $1.8 trillion in receipts. Solo founders' share of new startups more than doubled in a decade, from 17 percent in 2017 to about 35 percent in 2024. The number of solo operators clearing $100,000 a year is up roughly a third since 2022, and seven-figure solo businesses now number in the hundreds of thousands. Stripe's economists gave the category a name, the nanocorp, and noted that the median AI-native solo company sells into 55 countries in its first year. Sam Altman has described a betting pool among tech CEOs on the year the first one-person billion-dollar company appears. It has not happened yet, and the honest caveat is that the current solo ceiling is mid-eight-figure revenue. But the direction is not in dispute, and almost none of it is venture-funded. The solo boom is running on bootstrapped revenue, not other people's money.
Here is one point on that curve, in the founder's own words. Nevo David built Postiz, an open-source social media scheduling tool, alone, starting in September 2024. Scheduling tools are a twenty-year-old category with entrenched competitors, and for a year he ground it out to roughly $20,000 a month, respectable and unremarkable. Then he made one strategic decision: he rebuilt the product so AI agents could operate it. A public API, then a CLI so agents could drive it with short commands, then a skill file agents could discover, then an MCP server. He changed the homepage to say the quiet part: run your social media on autopilot with AI agents. Humans who wanted the old thing got a small button.
A stranger with 200 followers wrote an article about automating social media with agents through Postiz. It pulled 7.2 million views, because the demand was already there waiting for a product shaped correctly. This month Nevo posted the number himself: $145,000 in monthly recurring revenue, growing about $1,000 a day, about to cross $2 million a year. One person. The same technology wave that was shrinking Xbox was, at the same moment, letting a solo developer outgrow entire funded teams, and the difference was not talent or capital. It was that he sold an outcome to whoever, or whatever, showed up to buy it.
So the technology is not choosing sides. It is repricing a single input, execution, and the consequences land differently depending on what you own.
If you own an outcome, a product, an audience, a client relationship, a specification only you can write, AI is a force multiplier with no historical precedent. The gaps that used to force people to hire, the marketing you couldn't write, the code you couldn't ship, the books you couldn't keep, are now filled by tooling that costs less per month than a single recruiting fee. This is why the solo numbers bend upward right as the corporate numbers bend down.
If you own a seat, the honest news is worse, and pretending otherwise helps nobody. The safety that employment offered was always a bundle: income, identity, and the comfort of being needed. The third element is the one being repriced. For people who chose employment precisely because they value security, this is a frightening decade, and the fear is rational. The Xbox employees who lost their roles in July did nothing wrong. The arithmetic changed above them.
But the response to rational fear is not to burn tokens harder or grip the seat tighter. It is to notice what the new arithmetic actually rewards. The human does not disappear from Hölke's model; the human moves upstream, from operator to spec-writer and auditor. In career terms that means three things. Learn to find problems, not just solve them, because agents solve well-defined problems and someone still has to define them. Own the last mile, the judgment, taste, and accountability that separate an agent's median output from something a client will pay for. And build things that accrue to you: reputation, relationships, a product, a name, because those are the only assets that survive a restructure you don't control.
None of this requires quitting your job tomorrow. It requires refusing to let your entire economic identity be a seat in someone else's routing table.
The betting pool will eventually pay out. Someone, somewhere, will run a billion-dollar company from a laptop, and the milestone will get all the coverage. But the milestone is not the story. The story is the 30 million people already running companies of one, the solo operator adding a thousand dollars of recurring revenue a day by serving customers who are not even human, and the 1,600 people who opened a memo on a Monday morning and learned which side of the trade they had been standing on.
Software was never the safe career. Outcomes were.