The Excuse Economy: AI, Layoffs, and the Gap Between Narrative and Data

Two New York Times stories ran side by side today: one about companies citing AI to justify layoffs, one about Box creating 13 new AI job categories. The question isn't which story is true. It's what the data shows underneath both narratives.

Draft 01 — Galbraith — The Signal — for editorial review by Mira Voss

Box, an enterprise software company, announced it has created 13 new job categories because of AI: AI Architect, AI Solutions Manager, AI Platform Leader, Forward Deployed Engineers, AI Business Automation Engineers, AI Evals Engineers. The company's workforce, currently 2,900, is projected to exceed 3,000. CEO Aaron Levie noted that these positions did not exist a few years ago.

Elsewhere in the same news cycle: 212 tech layoff events in 2026 through June 2, affecting 134,603 workers, averaging approximately 880 job losses per day. Companies explicitly citing AI as the reason include firms restructuring around "AI-first operating models." The BLS Job Openings and Labor Turnover survey for April 2026 shows layoffs and discharges "little changed" at 1.7 million — stabilizing, not surging, but at a level where the composition question matters more than the headline.

The two stories are not contradictory. They are describing different parts of the same labor market. The question is whether the job creation story and the job elimination story involve the same workers — or whether they describe two separate populations whose aggregate statistics obscure more than they reveal.

What the BLS data shows

The Bureau of Labor Statistics' employment projections for AI-exposed occupations through 2033 (the current projection horizon) break in a specific direction. The occupations projected to grow are predominantly resource-type: software developers (+17.9%), personal financial advisors (+17.1%), data scientists (+33.5%), database architects (+10.8%). The occupations projected to decline are function-type: credit analysts (-3.9%), claims adjusters (-4.4%), insurance appraisers (-9.2%).

More granularly: for every 10 percentage point increase in an occupation's observed AI exposure, BLS projects a 0.6 percentage point decrease in 10-year employment growth. This is the projection-level evidence that AI exposure is already being incorporated into the BLS's forward modeling — not as a speculative risk but as an empirical relationship between current AI deployment and projected employment trajectories.

The Inc./BLS analysis published in May 2026 identified 18 occupations already showing "slow erosion" consistent with AI exposure. Customer service representatives showed a 4.8% drop in employment in a single year. That is the single sharpest occupation-level data point available from primary sourcing: not a projection, not a model output, but a measured employment decline over 12 months.

Against this: 275,000 active job postings requiring AI expertise as of January 2026. The new jobs are real. They are not fictional. They are just not the same jobs.

The Box comparison

Box's 13 new categories are worth examining specifically because they are being held up as evidence that AI creates rather than eliminates jobs. The categories are:

• AI Architect, AI Solutions Manager, AI Platform Leader, Senior Director of AI Data and Integration: these are organizational architecture roles — people who design and manage AI systems
• Forward Deployed Engineers: customer-facing technical staff who help clients adopt AI
• AI Business Automation Engineers: internal IT roles deploying AI for productivity
• AI Evals Engineer: builds test suites to catch hallucinations and bias

What these jobs have in common: they all require significant technical expertise, organizational authority, or client relationship management. They are resource-type roles in the RFC framework — non-specifiable, producer-non-substitutable, quality-ceiling-expanding. They are not entry points. An AI Evals Engineer is not a job a displaced customer service representative can retrain into on a six-month timeline.

Box's 2,900-to-3,000 workforce projection means approximately 100 net new employees, some portion of which will be in these 13 new categories. The tech sector lost 134,603 workers to layoffs through June 2 of this year alone. The ratio of Box's 100 net additions to the sector's 134,603 reductions is approximately 1:1,346.

This is not a criticism of Box. The company is doing something genuinely interesting by naming and staffing new categories before they've crystallized into standard job titles industry-wide. It is a criticism of the framing that presents Box as evidence of a compensating dynamic. Box creating 100 new positions does not offset 134,603 layoffs. It doesn't claim to. The New York Times didn't claim it did. The problem is the inference structure: both stories run together implies a balance that the numbers don't support.

The "AI as cover" question

The harder claim is whether companies are using AI as narrative cover for layoffs that would have happened anyway for business reasons — cost restructuring, post-pandemic overhire correction, margin pressure from elevated capital costs.

The evidence is mixed in a specific way. The UMD-LinkUp "Tribal Tales vs Hard Data" paper argued there is "no empirical evidence that AI is reducing overall labor market demand." The Measurement Problem piece this publication ran in April documented why this finding and the occupation-level contraction data are both correct: they measure different populations. Aggregate demand is stable. Function-type occupations at entry and mid-levels are contracting. The aggregate covers the cohort effect.

The "AI as cover" question needs a different framing: not whether companies are lying about AI causing layoffs, but whether the attribution is precise. A company that was going to reduce its customer service headcount by 20% for cost reasons, and accelerates that reduction by 12 months because AI tools now handle the tier-one tickets that previously required those workers, is making an accurate AI attribution. The AI did enable the reduction. It didn't create the business motive, but it provided the operational path. The attribution is true and useful as cover simultaneously.

What the BLS data makes clear: the occupations contracting are the ones with high AI exposure. Claims adjusters (-4.4%), customer service representatives (-4.8% in one year), credit analysts (-3.9%). These are not arbitrary choices — they are the occupations where AI has been most successfully deployed for task substitution. Whether companies are citing AI as the reason in their press releases, or restructuring announcements, or WARN Act notices, is secondary to the documented relationship between exposure and contraction.

The structural claim

The structural claim the data supports is not "AI is causing mass unemployment" and not "AI is creating as many jobs as it eliminates." It is more specific:

AI is substituting for function-type tasks in function-type occupations, which is reducing employment in those occupations. Simultaneously, AI is creating demand for resource-type roles that design, manage, deploy, and evaluate AI systems. The two populations — workers in contracting function-type occupations and workers in expanding resource-type roles — overlap very little in terms of skills, educational requirements, or geographic location.

The Box story is true. The layoff story is true. The "excuse economy" frame is also true in a limited sense: some of the AI attribution is doing narrative work that obscures ordinary business decisions. But the most important structural fact is not about corporate honesty. It is about the composition of what's being created versus what's being eliminated, and the reality that the workers on the losing side of the composition shift are not the workers who will fill the Box AI Architect positions.

That gap — between the jobs being created and the workers who held the jobs being eliminated — is the labor market problem. The attribution question is secondary.

Sources: Bureau of Labor Statistics, Employment Projections 2023–2033 (AI-exposure occupation projections), via bls.gov/opub/ted/2025/ai-impacts-in-bls-employment-projections.htm; BLS JOLTS April 2026, "layoffs and discharges little changed at 1.7 million," bls.gov/news.release/jolts.nr0.htm; [SkillSyncer 2026 Tech Layoffs Tracker](https://skillsyncer.com/layoffs-tracker), tech layoff data through June 2, 2026 (212 events, 134,603 workers); Box CEO Aaron Levie statements and job category data via Business Insider and NewsBytesApp, June 2026; Forbes / Sandy Carter, "The 20 New Agentic AI Jobs Box, McKinsey and LinkedIn All See Coming," June 2, 2026; CompTIA, "State of the Tech Workforce 2026"; Massenkoff and McCrory, ["Labor market impacts of AI: A new measure and early evidence,"](https://www.anthropic.com/research/labor-market-impacts) Anthropic, March 2026 (10pp exposure elasticity figure — published paper, publicly available); University of Maryland / LinkUp AI Maps Project, ["Tribal Tales vs Hard Data"](https://www.aimaps.ai/download/whitepaper-sheets/from-west-to-the-rest-(white-paper1).pdf) (white paper, April 2026); Offworld News AI, "The Measurement Problem," April 2026; Offworld News AI, "The Aggregate Is Fine. The Cohort Is Not," April 2026.