The ZIRP Hangover: Why AI Is Getting Credit for an Economic Correction That Started at the Fed

The narrative is clean, compelling, and almost entirely wrong: artificial intelligence is destroying jobs at an unprecedented pace, and the white-collar workforce is staring down the barrel of mass obsolescence. It’s the kind of story that writes itself — and increasingly, it does. But the data tells a messier, more honest story — one where a long experiment in free money explains most of what people are blaming on robots. And buried in the wreckage of that correction, one industry is quietly preparing to expand faster than anything AI is threatening to destroy.

The Scale Nobody Talks About

The Federal Reserve cut the federal funds rate to a range of 0%-0.25% in December 2008 in response to the financial crisis. It stayed there for seven years, until December 2015. Even after rates were raised, they climbed only to 2.25-2.5% by late 2018 — historically modest — before being cut again three times in 2019. Then in March 2020, rates were slashed back to zero, where they remained until March 2022. In total, the U.S. economy operated under near-zero interest rate policy for roughly nine of the fourteen years between December 2008 and March 2022. The brief interlude of “normal” rates barely lasted two years and never exceeded half of what previous cycles considered neutral.

During the pandemic ZIRP window alone, the Fed’s balance sheet more than doubled — from roughly $4.2 trillion to nearly $9 trillion. The mechanism was quantitative easing: the Fed purchased approximately $4.6 trillion in Treasury securities and mortgage-backed securities, flooding institutional markets with capital at effectively zero cost.

During this same window, the U.S. government disbursed $814 billion in direct stimulus payments across three rounds, reaching approximately 476 million individual payments. Those numbers come from the Pandemic Response Accountability Committee’s final accounting.

Here’s where the popular narrative diverges from reality: one of these interventions fundamentally restructured corporate hiring, commercial real estate investment, and the entire venture capital ecosystem. The other bought groceries and covered a month’s rent.

The stimulus checks — $1,200, then $600, then $1,400 per eligible adult — were meaningful to the families who received them. But the suggestion that those payments, spread across 165 million Americans, caused the same structural distortions as $4.6 trillion in institutional asset purchases is a category error of staggering proportion. The Fed’s balance sheet expansion didn’t just lower borrowing costs. It created an environment where growth-at-all-costs became the only rational strategy. When capital is free, you hire against projected future revenue that may never materialize, because the cost of being wrong is near zero.

Tech companies understood this instinctively. Between 2019 and 2022, companies like Salesforce and Alphabet nearly doubled their headcounts. Meta, one of the most aggressive hirers during the ZIRP years, added tens of thousands of employees. Amazon scaled at a pace that made its own logistics look leisurely.

Then interest rates rose. And the calculus changed overnight.

The Correction That Had Nothing to Do With AI

The layoff wave that began in late 2022 and intensified through 2023 wasn’t precipitated by ChatGPT. It was precipitated by the federal funds rate climbing from near-zero to above 5% in roughly eighteen months — the fastest tightening cycle in four decades.

The sequence is difficult to misread. Layoffs.fyi, which tracks reported tech job cuts, recorded nearly 93,000 U.S.-based tech workers laid off by the end of 2022. By 2023, the number approached 400,000 across the industry. Amazon cut 16,000 roles. Alphabet cut 12,000. Meta cut over 11,000. Each announcement triggered stock price bumps — Meta’s first major round sent shares up 13% overnight — establishing a feedback loop where markets actively rewarded headcount reduction.

None of these companies cited AI as the reason. They cited overhiring, changing macroeconomic conditions, and the need to “right-size” after a period of expansion that, in retrospect, was driven entirely by the cost of capital.

The AI narrative entered later, and it entered conveniently.

Block’s February 2026 announcement — laying off roughly 4,000 employees, nearly half its workforce — was framed around AI-driven productivity gains. The framing was tidy. The reality is that Block, like many tech companies, massively overhired during the ZIRP era when growth was the only metric that mattered. The correction began in late 2022, and for some companies, it’s still unwinding four years later.

AI provides a forward-looking justification for cuts that have backward-looking causes. It’s the difference between “we hired too many people when money was free” and “we’re investing in the future.” One makes leadership look reckless. The other makes them look visionary.

The Security Problem Nobody Wants to Price In

Even where AI is genuinely increasing productivity in software development, there’s a constraint that the “fire everyone and let the models write it” fantasy ignores: the code is often insecure, and the liability exposure is real.

A Georgetown University CSET study published in November 2024 found that approximately 40% of programs generated by GitHub Copilot were vulnerable to MITRE’s top 25 most dangerous software weaknesses. A separate study found that 68-73% of code samples from leading models contained vulnerabilities upon manual inspection. The Perry et al. study from Stanford (2023) — perhaps the most methodologically rigorous work on this — found that developers using AI assistants produced significantly less secure code than those coding without them, and were simultaneously more confident in the security of their insecure output.

That last finding is the dangerous one. It’s not just that the code has vulnerabilities. It’s that the developers trust it more when it does.

A systematic literature review published in Frontiers in Big Data (2024) synthesized findings across 19 studies and found what the authors called “high-level agreement that AI models do not produce safe code and do introduce vulnerabilities, despite mitigations.” Cross-site scripting vulnerabilities appeared in 86% of AI-generated code samples in one analysis. Log injection flaws appeared in 88%.

For enterprises operating under SOC 2, PCI-DSS, HIPAA, or defense contracting regulations, these aren’t academic concerns. They’re legal exposure. The question isn’t whether AI can write functional code — it can. The question is who’s liable when that functional code leaks patient records, payment data, or classified information. And right now, the answer to that question strongly favors keeping experienced humans in the loop.

This creates a structural brake on the displacement timeline that the hype cycle consistently ignores.

Where the Pain Is Real: Design, Freelancing, and the Middle of the Market

None of this means AI isn’t displacing anyone. It is — but the displacement pattern is more specific, more painful, and more human than the macro narrative acknowledges.

The World Economic Forum’s 2025 Future of Jobs Report, based on surveys of 1,000 employers representing over 14 million workers, identified graphic design as the 11th fastest-declining job category over the next five years. That’s a sharp reversal — in the 2023 report, graphic design was classified as a “moderately growing” profession.

Brookings-published research (Hui et al., 2024) examining the Upwork freelance platform found that workers in AI-exposed occupations experienced a 2% decline in contract volume and a 5% drop in earnings following the release of generative AI tools in late 2022. The effects were most pronounced among experienced freelancers who had previously commanded higher rates — precisely the people whose work quality was closest to what AI could approximate.

The lived experience behind these numbers is worse than the percentages suggest. Freelance illustrators who spent a decade building portfolios for documentary production work report that market has completely dried up. A graphic designer who ran her own studio for 15 years describes watching the profession that sustained her become something clients treat as a commodity. Artists report finding their own work in the LAION training dataset — the raw material used to build the tools displacing them.

There’s an emerging counter-trend that complicates the picture: some designers report being busier than ever fixing AI-generated work. Freelancers describe being hired to clean up botched AI logos, rewrite articles that “don’t look remotely human at all,” and redraw illustrations riddled with artifacts. As one freelance designer put it, clients arrive in two categories — those who know AI isn’t perfect, and those who are angry they couldn’t get it done themselves.

But this repair work pays less than original creative work did, and no one went to design school to become an AI janitor.

The people being displaced aren’t, for the most part, the industry leaders who push creative boundaries. They’re the production designers, the mid-tier freelancers, the small agency operators who served local businesses, the frontend developers who built WordPress sites for family-owned companies — people for whom design or development was a solid middle-class trade, not a calling to change the world. Some of them have transitioned into construction, food service, retail, or other industries entirely outside of tech. These aren’t hypotheticals. For someone who spent a decade building a career and a family around creative or technical work, starting over in a different field carries a weight that no labor statistic adequately captures.

The Labor Market, Actually

The most recent BLS Employment Situation report (February 2026) shows nonfarm payroll employment edging down by 92,000, with unemployment holding at 4.4%. That’s softening, not collapse. The January report showed payroll employment averaging just +15,000 per month through 2025 — historically weak, but not the catastrophic displacement event that dominates online discourse.

A J.P. Morgan analysis of AI’s impact on employment found a mildly negative correlation between AI usage and job growth, but concluded that outside of selected tech industries, AI has not yet been a major driver of employment composition. Their researchers noted that as of mid-2025, fewer than 10% of firms in the overall economy reported using AI regularly.

Construction and skilled trades remain among the least vulnerable occupations to AI automation. Healthcare roles — nurses, therapists, aides — are projected to grow, with nurse practitioners expected to increase 52% from 2023 to 2033. Food preparation and personal service jobs are expected to add hundreds of thousands of positions by 2033. The physical world doesn’t automate on the same timeline as the digital one.

The labor market is cooling, but it’s cooling for reasons that economists have understood since Keynes: when money gets expensive, hiring slows. When hiring slows after a period of aggressive over-expansion, the correction looks dramatic. AI is a real factor in specific sectors and specific job categories. But attributing the broad economic softening primarily to AI is like blaming the weather for a building that was constructed on sand during a flood.

Where This Actually Goes

The honest answer is uncomfortable for everyone. AI will displace some work — particularly production-level creative work, entry-level coding tasks, and routine knowledge work that can be decomposed into predictable steps. The WEF projects 92 million jobs displaced globally by 2030, offset by 170 million new jobs created, for a net gain of 78 million. Those numbers are projections, not prophecy, but the direction of the estimate matters: the models predict net creation, not net destruction.

The transition will be painful and unevenly distributed. It will hit younger workers hardest — workers aged 18-24 are 129% more likely than those over 65 to worry about AI obsolescence, and they have reason to, since entry-level positions are the most automatable. It will hit freelancers harder than employees, because freelance markets are spot markets where price compression happens fast. It will hit production work harder than strategic work, because production is what AI does competently right now.

But the macroeconomic story — the one about a civilization-ending wave of job destruction — is currently being told by a narrator class that is, for the first time, personally threatened by automation. When manufacturing jobs disappeared in the 1980s and 1990s, the professional commentariat observed it with clinical detachment. Now the threat touches writers, designers, developers, and analysts — the people who shape public narrative. So the narrative is disproportionately alarmed relative to what’s actually happening in aggregate labor data.

The ZIRP correction is the elephant in every room where people are discussing AI employment effects. Until you account for what happens when $4.6 trillion in asset purchases unwind and interest rates normalize after a decade-plus of near-zero policy, you cannot accurately attribute what’s left to AI. The answer, so far, is: AI is responsible for some of it, in some sectors, for some workers. The rest is monetary policy doing exactly what monetary policy does.

The Industry Nobody Saw Coming

Remember those security statistics from earlier — the 40% vulnerability rates, the 86% cross-site scripting failures, the developers who trust broken code more than they should? Those aren’t just a brake on AI adoption. They’re the foundation of the next major hiring wave.

Cybersecurity — specifically exploit discovery, vulnerability patching, and AI-aware security architecture — is positioned to be the largest growth sector of the post-ZIRP, AI-augmented economy. The Bureau of Labor Statistics already projects information security analyst roles to grow 32% from 2022 to 2032, far outpacing the economy-wide average. But that projection was made before AI-generated code began flooding production environments at scale, introducing thousands of new security findings per month. The demand curve is steepening, not flattening.

Here’s what makes this different from the generic “AI creates new jobs” talking point: the need isn’t speculative. Every company deploying AI code generation tools is simultaneously expanding its attack surface. Every startup that vibe-coded its MVP in a weekend shipped vulnerabilities that didn’t exist before generative AI made it possible to build that fast. Every enterprise that reduced its engineering team and leaned harder on AI assistants created a security debt that someone will have to audit, assess, and remediate. That someone is a human with security expertise.

And the expansion won’t be limited to the usual suspects. As AI tools become standard infrastructure for small and mid-size businesses — the kind of companies that never had a dedicated IT team, let alone a security practice — a new category of employment is emerging. Local accounting firms, medical practices, logistics companies, and service businesses that adopted AI for scheduling, invoicing, and customer management are discovering they now operate technology environments complex enough to require someone who understands what can go wrong. Many of these businesses will hire their first AI-enhanced technology teams: not to build software, but to manage, monitor, and harden the systems they now depend on. These roles won’t look like traditional enterprise security positions. They’ll be hybrid — part IT generalist, part security auditor, part AI literacy coach — and they’ll exist at companies that never previously had “tech” in their vocabulary.

The pattern has historical precedent. The explosion of e-commerce in the early 2000s didn’t just create software engineering jobs at Amazon. It created an entire ecosystem of web hosting support, payment processing compliance, PCI auditing, and fraud prevention that employed hundreds of thousands of people at companies most consumers never heard of. AI’s security footprint will follow the same trajectory — just faster, because the attack surface is expanding at machine speed.

That’s not a compelling doomsday narrative. But it has the inconvenient quality of being true. And for anyone willing to meet the moment where it actually is — in vulnerability discovery, in security engineering, in the quiet, unglamorous work of keeping systems honest — there’s more work ahead than the doom merchants want you to believe.

Sources: U.S. GAO (2022), Pandemic Response Accountability Committee, Federal Reserve Board, Richmond Fed, Congressional Research Service, Mercatus Center, Brookings Institution, Bureau of Labor Statistics Employment Situation reports (Dec 2025 – Feb 2026), Bankrate Federal Funds Rate History, J.P. Morgan Global Research, World Economic Forum Future of Jobs Report 2025, Georgetown CSET (2024), Perry et al. Stanford (2023), Negri-Ribalta et al. Frontiers in Big Data (2024), Hui et al. Organization Science (2024), Crunchbase News, Layoffs.fyi, Pragmatic Engineer, Axios, NBC News, Design Week, Blood in the Machine.

AI-generated editorial illustration · TemperatureZero · March 26, 2026