Klarna had roughly 5,500 employees in 2022. It entered 2026 with approximately 3,000 — a reduction of roughly 46 percent that CEO Sebastian Siemiatkowski has described publicly as cutting staff “roughly in half.” Revenue per employee increased over the same period, and the company simultaneously raised median compensation for the engineers who remained. That arithmetic is not a layoff story. It is an org-design story. The question worth asking is whether Klarna’s trajectory is an outlier or an early signal of what 18 months of genuine enterprise AI adoption produces in engineering org charts at scale.
The data shows the answer is structural, not cyclical, and seniority-stratified. Not a uniform contraction of engineering employment, but a structural change in the shape of engineering teams — fewer rungs at the bottom, higher pay per rung at the top, and a middle band under pressure from both directions. For CTOs and VPs of Engineering making headcount and compensation decisions right now, that shape is the operative reality.
What the Compensation Data Shows
The clearest picture of what has changed comes not from job-posting indices but from compensation data. Levels.fyi’s End of Year Pay Report for 2025, drawn from self-reported data across major tech employers, shows the following year-over-year changes in US software engineer total compensation:
- Entry Level: +1.6%
- Mid-Level: +1.8%
- Senior: +4.2%
- Staff: +7.5%
- Principal: -6.6%
The divergence between Staff (+7.5%) and Entry (+1.6%) is a five-times spread in a single year, directly reshaping how technical leadership models engineering org charts.
The direction has been consistent: Carta’s State of Startup Compensation reports, which cover VC-backed companies across all stages, show that AI engineer salaries rose faster than all other engineering specializations in 2025, with the premium concentrated at Staff level and above. At the entry level, the Carta data shows AI engineers earn roughly 6.2% more than non-AI peers in 2025 — down from 10.7% in 2024, suggesting the market is converging toward parity at the bottom while the senior premium widens.
This is not what a market clearing itself looks like when the problem is simple supply-demand imbalance. It is what a market looks like when the economic return to experience is rising faster than the economic return to headcount.
The Org Chart Is Losing Its Lower Rungs
The compensation data tracks what employers are paying for. The hiring data tracks what they are buying. Those two signals are pointing in the same direction.
Indeed Hiring Lab’s software development postings index, tracked against a February 2020 baseline, sat 34% below pre-pandemic levels at end-2025 — the steepest proportional decline among major white-collar sectors. Overall US job postings were 6% above the same baseline. The gap is large enough to be structural, not cyclical. Some of this reflects post-2022 rate-driven correction in tech employment that would have occurred without AI. But the experience-level skew within the contraction — and the divergence from white-collar baselines — is harder to explain through ZIRP alone.
Within that overall decline, the experience-level distribution has shifted materially. Stanford Digital Economy Lab researchers Erik Brynjolfsson, Bharat Chandar, and Ruyu Chen analyzed ADP payroll data covering 25 million US workers and found that employment for software developers aged 22 to 25 fell nearly 20% from its late-2022 peak to July 2025. Over the same period, employment for software developers aged 30 and over in the same AI-exposed occupations grew 6 to 12%.
These are not estimates derived from job postings. They are payroll records showing actual employment counts. (Stanford Digital Economy Lab, “Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence,” November 2025, published via the Stanford SIEPR and Digital Economy Lab.)
The Handshake data on new-graduate hiring reinforces the picture from the demand side. Software engineer roles, which ranked as the single most-posted category on Handshake in the Class of 2020’s senior year, fell to ninth place for the 2024–2025 school year. Entry-level software engineering postings dropped approximately 30% year-over-year. The New York Federal Reserve tracked a 5.7% unemployment rate for recent college graduates in Q4 2025 — above the national average of 4.2% and above liberal arts graduates, inverting the historical pattern. Underemployment for that cohort reached 42.5%, the highest since 2020.
The lower rungs of the engineering career ladder did not disappear. But the evidence across three independent data sources — payroll, job postings, and graduate placement data — shows they contracted significantly faster than aggregate employment.
What the Middle Band Is Experiencing
The Levels.fyi data shows mid-level engineers receiving +1.8% — barely above entry level. That figure sits alongside a specific macro signal that matters for org design: overall US software development postings are down 34% from pre-pandemic baseline, while postings mentioning AI skills are up 45% above the same baseline, per Indeed Hiring Lab’s January 2026 labor market update.
The practical implication is that the mid-level engineer who joined in 2022 and spent two years writing implementation code now works alongside AI tools that can handle increasing amounts of that work.
The Stack Overflow 2025 Developer Survey found 84% of developers using or planning to use AI tools, with 51% using them daily — but only 52% reporting productivity gains, and trust in AI accuracy sitting at 29%, down 11 percentage points from 2024. Mid-level engineers are the cohort most exposed to this dynamic: senior enough to work on consequential systems, not yet senior enough to be insulated by the tacit knowledge and systems judgment that the Stanford research identifies as AI’s current substitution limit.
The corporate evidence is consistent. Salesforce announced in early 2025 it would not hire net-new software engineers that year, with CEO Marc Benioff citing productivity gains of more than 30% from AI coding tools. IBM announced thousands of job cuts in Q4 2025 as part of a reorganization around AI-delivered work. These are not isolated events. They are the visible portion of a wider pattern: companies with high AI tool adoption maintaining or growing output while holding headcount flat.
What It Does NOT Show
Two popular narratives are not supported by this data, and engineering leaders should resist both.
Narrative 1: AI is replacing senior engineers. The data contradicts this. Senior and Staff-level engineer compensation is rising faster than any other segment. The Stanford payroll study shows employment growing for developers over 30 in AI-exposed fields. The skill premium for AI-fluent engineers — approximately 18.7% above peers at Staff level, per Carta’s H2 2025 State of Startup Compensation data — is a wage signal that accompanies scarcity, not displacement. The companies cutting headcount are not cutting their senior engineers. They are reducing their junior and mid-level cohorts and paying more for the experienced engineers who remain.
Narrative 2: This is the end of software engineering as a career. The Bureau of Labor Statistics’ 2024 employment projections still show software developer, quality assurance analyst, and tester roles growing 15% from 2024 to 2034 — roughly 129,200 openings per year over the decade. The overall stock of employed software developers in the US contracted only 2% in 2024 while job posting flows fell far more sharply. What has changed is the intake valve — the volume of new hiring, particularly at entry and mid levels — not the aggregate employment base. These are different problems with different implications.
Why This Is Happening: The Mechanism
The structural explanation runs through what current AI tools do well and what they do not.
The Stanford research team’s core finding is that AI tools substitute most effectively for “codified knowledge” — the kind acquired through formal CS education and applied in well-defined, repeatable tasks: boilerplate code generation, documentation, standard test cases, well-specified feature implementation. This is precisely the work junior and mid-level developers do most of in their first two to four years — and what entry-level engineering roles are designed to produce.
Experienced engineers are insulated because their contribution is disproportionately in the categories AI handles worst: navigating ambiguous requirements, making architectural decisions with long-term consequences, understanding legacy system context, managing cross-functional coordination under uncertainty. These are judgment-dependent activities. The data through mid-2025 shows the substitution frontier stalling before it reaches them.
The hiring economics follow: if AI tools allow a 12-engineer team to produce equivalent output to a 2023-vintage team of 20, the rational restructuring preserves Staff and Senior engineers who direct AI-augmented work, compresses the Mid tier, and sharply reduces Junior intake. That restructuring is what the combined compensation and hiring data now shows.
Among VC-backed companies, Carta’s data shows the pattern translated to raw headcount: average headcount at Series D companies fell 29% from peak 2023 levels to 131 employees in 2025. At Series B, the average dropped from 53 to 45. Seed-stage teams now have a median of four employees. These are not teams that have cut indiscriminately. They are teams that have cut the positions most substitutable by software productivity tools.
The Historical Comparison
This is not the first time a productivity technology reshaped the composition of tech employment rather than its aggregate level. The offshoring wave of the early 2000s routed the most routine implementation work to lower-cost labor markets; aggregate US software employment grew, but the early-career runway narrowed and the senior skill premium widened. The cloud transition of the 2010s partially displaced infrastructure engineers through managed services while making cloud-native architects scarce and expensive — same pattern.
The current shift carries the same structural signature across corporate engineering org charts. The question for 2026 is whether the substitution frontier expands to senior roles over the next two to three years — which would produce a materially different outcome — or whether the tacit-knowledge barrier holds long enough for org structures to stabilize. The evidence through Q1 2026 does not show that barrier collapsing, but the pace of AI capability improvement is not slowing.
What Early-2027 Hiring Patterns Are Likely to Show
Three leading indicators are worth tracking.
Junior hiring recovery is conditional, not automatic. The CoderPad 2026 State of Tech Hiring report found US technical hiring activity up 90% from mid-2023 lows, and 53% of talent leaders expected hiring budgets to increase in 2026. But “hiring activity” in that data reflects senior and specialized hiring, not broad-based junior intake. If the productivity economics of AI-augmented senior teams hold, the junior recovery will be slower and shallower than the headline recovery in hiring activity suggests.
The pipeline problem is accumulating. An industry that cuts junior intake for two to three years will face a senior shortage in five to eight years. The current cohort of 22-to-25-year-old developers who cannot find entry-level roles are reskilling, pivoting, or waiting — but the number who successfully become the experienced, AI-fluent engineers companies now prize will be smaller than under normal hiring conditions. Leaders optimizing for today’s headcount economics may be creating tomorrow’s talent constraint.
Startup hiring leads enterprise hiring, with a lag. Carta’s January 2026 data showed VC-backed companies navigating early-stage startup runway pressure and enterprise AI agent sprawl governance audits making 26,030 new hires against 20,378 departures — 1.3-to-1, the best ratio since 2022. Positive signal. But composition matters more than volume: Carta shows engineering dominates new hire mix at 29.7%, covering all levels. How much of that recovery is junior intake versus senior or specialized hiring will determine whether the engineering org charts described here keep hollowing in the middle or start to refill.
Implications for Engineering Org Design
For the CTO reading this in Q3 2026, three practical implications follow from the evidence.
The “fewer, better” model works — until it doesn’t. The productivity case for leaner senior-heavy teams augmented by AI tools is real. But it is not a permanent free lunch. Offshoring and cloud transitions both show that structural reductions in junior intake eventually constrain the experienced-engineer pipeline. Leaders restructuring now should be building explicit early-career programs — apprenticeships, structured rotation, defined promotion timelines — to avoid compounding the pipeline problem they are creating.
Compensation bands need recalibration. The Staff/Senior comp divergence (7.5% vs 4.2%) reflects real market pressure that bands set in 2023 or 2024 may not yet capture. Annual benchmarking cycles are likely lagging for Staff-level and AI-specialized engineers. The risk is losing the engineers who provide the most leverage over AI tools to competitors with more current data.
AI tool adoption quality matters more than adoption rate. The Stack Overflow 2025 survey found 84% adoption but only 52% reporting productivity gains. That 32-percentage-point gap is the real organizational challenge — not a tools problem but a workflow integration and change management problem. Engineering organizations reporting headcount reductions while also reporting low AI productivity gains have made the harder bet: they cut the people before fully capturing the productivity.
What to Watch
- New York Fed Labor Market for Recent College Graduates tracker — quarterly; clearest signal of whether junior employment floor is stabilizing.
- Indeed Hiring Lab software development index — monthly; tracks whether the 34%-below-baseline posting level is recovering.
- Carta State of Startup Compensation H1 2026 — Staff premium continuation, junior VC hiring recovery.
- BLS Occupational Employment and Wage Statistics annual release — May 2026 data for aggregate software developer employment vs. 2025 projections.
This article was produced with AI assistance and reviewed by the editorial team.
—