On July 8, 2026, Indeed Hiring Lab published a report by economist Guillermo Gallacher that breaks a three-year pattern almost every AI-and-jobs headline has repeated. Occupations most exposed to AI, the same ones that lost the most job postings between 2022 and 2026, are now recovering faster than the rest of the labor market. US software development postings are up almost 15% since Claude Code launched in late February 2025, while total postings on the platform fell 7% over the same stretch. That’s a 22-point gap opening up in about a year and a half.

The number that matters more than the headline

Strip away the framing and look at what actually shifted. Between May 2022 (the post-pandemic hiring peak) and May 2026, the relationship between AI exposure and job postings was exactly what you’d expect from the coverage of those years: the more exposed an occupation was to AI, the more its postings fell. Indeed Hiring Lab calls this relationship statistically significant, and it lines up with prior research showing declines in AI-exposed vacancies started even before ChatGPT’s late-2022 release.

Then the report isolates the more recent window, May 2025 to May 2026, and the relationship inverts. The more exposed an occupation is to AI, on average, the more it rebounded. Not just software development. Other AI-exposed occupations show the same shape. The trend also isn’t US-only: the share of postings that are software development jobs is climbing in most large economies Hiring Lab tracked, with the exception of Germany and France, and English-speaking countries show the most consistent upward trend.

Indeed Hiring Lab says the honest thing twice in the same report: correlation doesn’t prove causation. There were plenty of forces moving the labor market between 2022 and 2026 that have nothing to do with AI: interest rates, the hangover from 2021’s overhiring, normal tech-sector cycles. But the report calls the timing overlap between Claude Code’s release and the start of the rebound “a coincidence that cannot be ignored,” and that’s a fair way to hold two true things at once.

The rebound has a ceiling most summaries skip

Here’s the part getting cut from most write-ups of this report. The recovery is not evenly distributed. 71% of the increase in US software development postings between May 2025 and May 2026 came from senior roles. 37% came from postings that mention AI directly in the title, and the two categories overlap heavily. Despite the rebound, software development postings remain roughly 27.5% below pre-pandemic levels, while overall postings are back to essentially where they were in February 2020.

Indeed Hiring Lab names the risk directly: this could still be a “seniority-biased technological change.” AI is amplifying experienced professionals who already have judgment to apply it to. It is not, at least not yet, reopening the entry-level door that’s been closing since 2022. That nuance matters for anyone reading this as unambiguous good news for the broader software labor market.

Dawson’s framing, kept as framing

Futurist Ross Dawson amplified the Indeed data on LinkedIn on July 12: “A fundamental shift. AI-exposed jobs were going down relative to others. Over the last year they are increasing substantially faster. AI is not replacing workers. It is resulting in organizations needing more knowledge workers.” He then listed six candidate explanations and prefaced them honestly: “we of course don’t know what will happen from here, or why this shift is happening.”

Two of his explanations are worth sitting with, treated as hypotheses rather than conclusions. One: leaders learned the expensive way, after firing and then scrambling to rehire, that AI complements most workers rather than replacing them. Two: AI capability crossed a real threshold, moving from an interesting but error-prone tool to something that genuinely amplifies an experienced professional’s output, which is a different investment case than the one companies were making in 2024 and early 2025.

The bet that was already on the table

Back in March, I wrote about Andrej Karpathy’s AI exposure map, which scored 342 Bureau of Labor Statistics occupations 0-10. I closed that piece citing Kevin A. Bryan, an economist at the University of Toronto, who publicly bet $1,000 that most “most susceptible” occupations would see a greater share of the labor market by 2030, not a smaller one. His logic: when the cost of the tasks a worker performs falls, demand for the full service that includes those tasks tends to rise.

That bet predates Indeed Hiring Lab’s report by months. I’m not citing it in convenient hindsight. It’s evidence that the mechanism Bryan proposed is already showing up in actual hiring numbers, not just in economic theory.

There’s also a direct line to Gartner’s May 2026 survey, which I covered when Standard Chartered’s CEO said AI would replace “lower-value human capital.” Gartner found that 80% of large organizations piloting AI report headcount cuts, with zero measured correlation to ROI. Klarna’s 2024 cut of 700 support roles, reversed in 2025 once the tool failed on complex cases, was the anecdote. Indeed Hiring Lab’s rebound in AI-exposed hiring is the first aggregate signal that the same shape might be playing out across enough companies to move a national number. That’s a correlation stacked on a correlation, not proof of a mechanism, and I want to be honest about that limit rather than oversell it.

A pattern I’ve watched before

I’ve been in this industry for 36 years, since I was writing code on a Commodore 64 in 1990. I’ve watched a version of this cycle repeat with enough regularity that I don’t need Indeed’s data to believe it, though it’s good to have it: a new tool arrives, a wave of companies cuts headcount betting the tool closes the gap by itself, the tool turns out to amplify rather than replace, and eighteen to twenty-four months later those same companies are rehiring into the same function, usually at a higher rate because the role now also requires knowing how to run the tool. I watched a version of it during the ERP consolidations of the 2000s. I’m watching a faster version of it now with agentic AI.

The senior-heavy shape of this rebound also tracks with what I found digging into Stanford/ADP payroll data on engineering hiring: junior developer employment down roughly 20% since 2022 while senior roles held steady or grew. Indeed’s 71%-from-senior-roles number and that bifurcation aren’t a coincidence. They’re the same labor market described from two different data sources.

None of this means your company’s cut corrects itself automatically. It means the question worth asking before signing off on the next “AI-enabled” reduction is whether you’re cutting capacity you’ll be rehiring in eighteen months, at a higher price, with fewer people left who remember why it was cut in the first place.

At IQ Source, AI Maestro’s discovery phase exists to answer that question with your own process data before the board forces a guess. A Process Reality Map and an AI Opportunity Score scored task by task, not by job title, so the amplify-or-cut decision gets made on evidence instead of the general promise that “AI handles it,” which has already gotten expensive for more than one company this year.

Map what to amplify before you cut