Citadel Securities published a macro strategy report last week that included a chart nobody on AI Twitter wanted to see: software engineer job postings on Indeed are climbing, up 11% year-over-year. Not flat. Not declining gracefully. Rising.
The report, authored by macro strategist Frank Flight from Citadel's London office, was a direct response to a viral Substack essay by Citrini Research that imagined a 2028 economic collapse driven by mass white-collar AI displacement. That essay spooked markets badly enough that Visa, Mastercard, and DoorDash shares all dropped on the Monday it spread. Michael Burry amplified it on X. Bloomberg covered the selloff.
Then Citadel showed up with actual labor market data.
The chart that complicates everything
Flight's core move is simple: pull Indeed job posting data, track it against overall postings, and let the trend speak. Software engineering postings have diverged upward from the broader market. The 11% figure is year-over-year as of early 2026, sourced from Indeed's daily and 21-day moving average data. It lands in a labor market where U.S. unemployment sits at 4.28% and AI capital expenditure has hit roughly 2% of GDP, about $650 billion.
Now, there's some context this chart doesn't give you. Job postings aren't hires. A posting that sits unfilled for six months still counts. And Indeed data tracks volume, not quality or seniority. The 11% bump could reflect a surge in senior roles while entry-level positions continue to shrink, which is exactly what other data suggests. Entry-level software engineering postings are down roughly 40% from pre-2022 levels, according to multiple industry reports. So "more jobs" might really mean "more jobs for people who already have jobs."
Flight doesn't engage with that nuance much. His argument is broader: the doom narrative doesn't match the macro data, full stop.
Why Citadel thinks the panic is wrong
The report pulls in St. Louis Fed data from the Real Time Population Survey to make a second point: daily AI usage for work isn't accelerating. If AI were about to gut white-collar employment, you'd expect to see a sharp uptick in how intensely people use it at work. The data, Flight argues, is "unexpectedly stable" and shows little evidence of near-term displacement risk.
His framing is pointed. Citrini's essay projected a 10.2% unemployment rate by 2028 and an S&P 500 drawdown of 38%. Flight's response, paraphrased: macroeconomists can barely forecast payroll numbers two months out, but a Substack post can apparently map the destruction of the labor market with confidence.
"Recursive capability does not imply recursive adoption," the report states, and this is where the argument gets interesting. Even if AI models keep improving, economic deployment hits real walls: energy costs, data center capacity, regulatory friction, organizational change. If compute demand spikes faster than supply, the marginal cost of automation rises above the marginal cost of human labor, and substitution just stops. There's a natural ceiling that the displacement narratives tend to ignore.
Jevons and the demand question
The people sharing Citadel's chart on social media have latched onto a specific framing: Jevons Paradox. The idea, from 19th-century economist William Stanley Jevons, is that when a resource becomes cheaper to use, total consumption goes up rather than down. More efficient steam engines didn't reduce coal usage. They made coal-powered activity so cheap that Britain burned far more of it.
Applied to software: if AI makes writing code dramatically cheaper, companies don't stop hiring developers. They start building software for use cases that were previously too expensive to justify. Internal tools for niche operations. Custom automation for small businesses. Software in industries that never had dedicated engineering teams.
It's an appealing framework, and there's some historical backing. Cloud computing, open source libraries, better tooling: all of these made software development more efficient over the past two decades, and the industry kept growing. As HackerRank noted in a recent analysis, earlier productivity leaps in development have consistently raised expectations rather than reduced headcount.
But Jevons doesn't always hold. Farming is the classic counterexample: productivity per laborer skyrocketed over the last century, and farm employment collapsed from a third of the U.S. workforce to about 1.3%. Demand for food only stretches so far. The question for software is whether demand is more like coal or more like food. Most people making the Jevons argument assume demand for software is effectively infinite. That's a strong assumption, and the Citadel report doesn't really test it.
What the report doesn't address
Flight invokes Keynes's 1930 prediction that productivity growth would shrink the workweek to fifteen hours. Keynes was right about productivity, wrong about what people would do with it: they consumed more rather than resting more. Flight sees the same pattern with AI. Productivity gains expand the consumption frontier, and labor adapts.
Maybe. But the Citadel report is a snapshot, and snapshots can be misleading. The 11% increase in job postings could reflect companies staffing up for AI integration projects that are themselves temporary. Data center construction is booming, which the report acknowledges is boosting construction hiring. Once those centers are built, those jobs vanish.
The report also makes no attempt to reconcile its optimism with the very real pain in entry-level hiring. If you're a recent CS grad, the macro data is cold comfort when junior developer postings have cratered. IBM's HR chief has said AI will create "totally different jobs," with entry-level coders moving toward marketing, client relations, and product work. That's a polite way of saying the bottom rungs of the engineering ladder are being sawed off.
Citadel's report runs about 2,500 words and references roughly $650 billion in AI capex, 2,800 planned U.S. data centers, and a 4.28% unemployment rate. It frames AI as a classic positive supply shock: disinflationary, growth-enhancing, historically comparable to steam power and electrification. The framing is deliberately calm in a market rattled by Citrini's sci-fi scenario.
Whether the calm is warranted depends on something neither report can prove: how fast adoption actually moves from here. Citadel bets on the S-curve flattening. Citrini bets on it going vertical. The Indeed data, for now, sides with Citadel. But "for now" is doing a lot of work in that sentence.
