The AI Jobs Reset

Every previous technology transition produced both job loss and job creation. Every one. The historical record on this question is not contested — the assembly line eliminated entire crafts and created entire factories; the personal computer hollowed mid-tier clerical work and built a software industry from nothing; the mobile internet displaced retail employment categories that had absorbed two generations of workers and seeded a creator economy that has now absorbed two generations more. The pattern is so regular that the absence of the pattern would itself be the surprise.

What the steward needs to read is not whether AI will eliminate jobs — it will, the visible side of the ledger already records 125,098 of them in the United States since January 2025 — but whether the creation side of the ledger is moving with the destruction side or behind it, and by what margin.

The destruction side is easy to count. The Alliance for Secure AI tracks every public report against three attribution tiers. The cumulative number rises in a clean curve. The Technology sector accounts for about 90% of it today — which is itself the most important fact in the data, because if 90% of AI-driven layoffs are still inside the sector that builds AI, the substitution dynamic has not yet crossed into the rest of the economy. When that share falls below 75%, the conversation changes structurally.

The creation side is harder to count, and that is the reason most analyses get this question wrong. AI-native job categories — Prompt Engineer, AI Engineer, MLOps Engineer, AI Trainer, AI Risk Officer — emerge faster than the BLS can codify them. Job posting platforms catch the early signal: postings explicitly titled with these roles are up materially year-over-year, and the search-interest signal for "prompt engineer" leads posting volume by two to three quarters. Carta data on private AI companies shows headcount growth running ahead of the public-company displacement at the venture-stage cohort level.

The historical analogue is not the textile mill. It is the personal computer. By 1985, "secretary" was a vanishing job title and "computer programmer" was a category most parents could not yet name. The displacement was real and concentrated; the creation was real and diffuse. The visible side of that ledger dominated news cycles for a decade. The diffuse side won the count in the end.

What this Current measures is the running net — the gain-side signals weighed against the loss-side signals — so that the steward holding a workforce, a real-estate footprint, or a hiring plan can read whether the labor market is balancing or shrinking, and by how much, before the political and policy surfaces resolve.

The destruction-side reads are not wrong. They are accurate, traceable, and load-bearing. They are also one half of an arithmetic.