Snap cut 1,000 employees on April 15 and told investors that 65% of its new code is now AI-generated. The stock jumped 7%. CEO Evan Spiegel’s memo cited “rapid advancements in artificial intelligence” as the reason smaller teams could handle work that previously required larger ones. Annualized savings: over $500 million.
Twelve days earlier, the Federal Reserve published a FEDS Note tracking AI adoption across the U.S. economy. The headline: about 18% of firms have adopted AI as of year-end 2025. The workforce-level number is higher — 41% of workers report using generative AI — but aggregate productivity gains remain statistically invisible.
Then on April 14, an Atlanta Fed and Richmond Fed working paper surveying nearly 750 corporate executives found something sharper: a “productivity paradox” where perceived gains outstrip measured ones. Executives feel more productive. Revenue data doesn’t confirm it. The authors attribute the gap to “a delay in revenue realizations” — polite Fed-speak for: the money hasn’t shown up yet.
This matters for the SaaS hypothesis because the entire seat-compression thesis depends on a chain of logic: AI makes workers more productive → companies need fewer workers → fewer workers need fewer software seats → SaaS revenue per customer shrinks. If the first link is weaker than assumed, the chain doesn’t break — but it bends in uncomfortable directions.
The Numbers That Don’t Add Up
Q1 2026 brought 78,557 tech layoffs, with 47.9% directly attributed to AI. Snap. Oracle. Block. Atlassian. The rhetoric is uniform: AI enables us to do more with less.
But Goldman Sachs and JPMorgan both cut their 2026 U.S. productivity growth forecasts from 2.5% to 1.8% annually. The Fed’s own survey found that 68% of firms report AI productivity gains of just 0–5% — roughly equivalent to a standard software upgrade. Nearly 90% of executives across a global survey said AI had “no measurable impact” on employment or productivity over the past three years.
Economists have a name for this: the AI Solow Paradox, after Robert Solow’s famous 1987 observation that “you can see the computer age everywhere but in the productivity statistics.” The parallel is uncomfortably exact. Companies are spending hundreds of billions on AI infrastructure. They’re firing people and citing AI as the reason. But the productivity data is flat.
Two Explanations, Both Bad for SaaS
One possibility: the productivity is real but delayed. The Atlanta Fed paper leans this way — the “J-curve” hypothesis, where firms eat implementation costs before gains materialize. Goldman Sachs still projects AI will add 0.4 percentage points to productivity growth later in the decade. If this is right, the seat compression is coming but hasn’t fully arrived. Current layoffs are partly anticipatory — companies cutting preemptively, collecting the market reward now (Block’s stock jumped 6% on its 40% headcount cut), and betting the AI capability catches up.
The other possibility: AI is a convenient justification for cuts that would happen anyway. Irenic Capital pressured Snap before the AI-attributed layoffs. Oracle’s 30,000-person cut coincides with a $156 billion capex pivot, not a productivity breakthrough. The “AI made us do it” framing gives management cover for restructuring that activist investors and market incentives demanded regardless.
For SaaS vendors, both paths lead to the same immediate outcome — fewer seats sold — but with very different medium-term implications. If the productivity is real and delayed, seat compression accelerates as gains compound. If it’s narrative-driven, the cuts may overshoot, companies may rehire, and seat counts could stabilize or rebound. The Atlanta Fed’s “compositional reallocation” finding — routine clerical roles declining while skilled technical roles increase — suggests the answer may be both simultaneously: real displacement in some job categories, performative in others.
What the Fed Actually Found
The detail that matters most sits in the Atlanta Fed paper’s sector analysis. Labor productivity gains from AI are “concentrated in high-skill services and finance.” Not uniformly distributed. Not hitting every SaaS customer equally.
This maps to a pattern already visible in public market data. ServiceNow reports Q1 2026 earnings tomorrow (April 22), and the market is watching whether its Now Assist AI suite — targeting $1 billion ACV — can deliver subscription revenue growth of ~21% despite a 72% year-over-year contraction in federal obligations. The companies that sell to high-skill services and finance might see more usage as those sectors adopt AI. The companies that sell seat-based licenses to the routine clerical roles being eliminated? They’re the ones staring at the structural problem.
The Fed data suggests the seat-compression mechanism is real but narrower than the market narrative implies. It’s not “AI replaces all workers who use software.” It’s closer to: “AI eliminates specific categories of routine work, and SaaS products priced per human doing that work lose those seats.” The difference between a uniform compression story and a targeted one is the difference between a sector-wide crisis and a sorting event.
Hypothesis Update
The Solow Paradox data complicates the hypothesis in a specific way. The hypothesis claims software loses value because AI creates it fast and efficiently. The Fed evidence says: AI adoption is widespread but AI productivity is unmeasured at scale. The seat compression happening now is partially real (concentrated in routine roles), partially anticipatory (companies cutting for market reward), and partially performative (activist-driven restructuring wearing an AI label).
This doesn’t falsify the hypothesis — the seats are genuinely disappearing. But it weakens the assumed mechanism. If the productivity doesn’t materialize at the pace priced in, we should expect a correction: some rehiring, some seat count recovery, and a revision of the timeline. The question to watch isn’t whether AI displaces workers — the Fed confirms compositional reallocation is happening — but whether the pace matches what SaaS multiples have already priced in.
ServiceNow’s earnings tomorrow will be one data point. The real test comes later: whether the companies that cut 16–40% of their workforce in Q1 actually ship more product with fewer people, or quietly backfill by Q4.
Sources: Federal Reserve FEDS Notes (Apr 3, 2026), Atlanta Fed/Richmond Fed Working Paper (Apr 14, 2026), Snap Inc. investor update (Apr 15, 2026), Goldman Sachs/JPMorgan productivity forecast revisions, Tom’s Hardware layoff tracker.