WIRED published an article on August 31, 2026, about the most anti-AI profession on the American labor market: according to Glassdoor data, insurance adjusters — employees who assess damage and claim amounts for insurance cases — leave negative reviews about AI in 98% of cases. Their pessimism is understandable: the industry is rapidly shrinking, the routine of initial claim intake and damage assessment is already performed by algorithms, and the cost of these algorithms' errors falls on people. This profession clearly shows how the "second wave" of corporate AI works: automation does not so much eliminate labor as redistribute it.

image

What happened

The WIRED article collects measurable traces of the profession's contraction. As late as 2024, the U.S. Bureau of Labor Statistics (BLS) forecast a minus of 18,900 jobs, or minus 5%, over a decade; reality turned out to be harsher — from May 2025 to May 2026, sector employment fell by 21%. The number of entry-level vacancies has decreased by 50% since 2025. The article names specific mechanisms of displacement: chatbots take over initial claim intake, analysis of photos and videos forms claim amounts without a specialist's visit, and AI summarization of medical records suffers from hallucinations. Startups are fueling automation: Liberate raised 50 million dollars for reasoning AI agents for insurance, and Pace — 46 million dollars for "agentic workforce." The ranking in which adjusters took last place in their attitude toward AI is built on Glassdoor review data.

Context

Insurance has become the first measurable market where vertical AI agents have passed the pilot stage and turned into paid demand. The case of insurtech company Lemonade is telling: its chatbot AI Jim processed 96% of initial inquiries and about 55% of all claims fully automatically by the end of last year. The work scheme is confidence-based routing: simple segments go on autopilot, complex cases are escalated to a human. The Liberate and Pace rounds formed a vendor ecosystem of "agentic workforce," and insurers have a buyer who is already trained to pay for automation. At the same time, the classic profession ladder is breaking down, where a junior started with routine: when entry through entry-level positions narrows, the channel of expertise transfer from experienced specialists to newcomers collapses — and it is on this basis that the ability to check others' decisions rests.

Why this matters for the industry

For the industry, this is a portrait of the "second wave" of corporate AI: implementation does not reduce labor, but redistributes it. Savings on visits and initial intake turn into a queue of manual rework — reclassification of claims and correction of hallucinations in summaries — and the reputational costs of errors fall on frontline employees, not on the model vendor. For builders, the signal is double. On the one hand, the buyer is formed and pays, but competition in the initial intake layer is already dense. On the other hand — value shifts to the seam between the AI solution and the human: review queues with filters by error types, QA panels for classification and hallucinations, routing engines by model confidence, and integration middleware between agents and insurers' legacy systems. In procurement, metrics will become increasingly important: share of classification errors, frequency of hallucinations, cost of disputed cases, and speed of escalation to a human. An important caveat about the defensibility of such products: without access to the insurer's data and pipelines, quality control tools turn into a thin wrapper that Liberate or Pace themselves can replicate if they wish — sustainability here is determined by integration and distribution. If the decline in junior hiring persists, the industry will face a shortage of people capable of verifying agents' decisions, and the cost of human verification will rise.

Why this matters for users

For the reader, the material is a clear instruction on how AI is implemented from the top and without the consent of the performers: when a model makes a mistake in claim classification, the person on the frontline suffers, not the vendor. As an insurance client, it is useful for you to know that a payout may be calculated by an algorithm from photos in seconds and without an assessor's visit. In case of a disputed decision, it is worth complaining about the final calculation and demanding a human check — especially if the case involves medical documents, which AI retells with hallucinations. Those who are considering the profession of an adjuster should take into account that entry through junior positions has noticeably narrowed, and the role itself is shifting to handling exceptions and appeals.

What is not yet known / limitations

Key figures — the share of negative reviews, the decline in employment and vacancies, Lemonade's automation metrics — are cited according to WIRED, Glassdoor, BLS, and the companies themselves and were not independently verified in this material. Reviews on Glassdoor are self-selective: they are written by those who decided to speak out, so 98% is a snapshot of the profession's mood, not a strict sociological measurement. It does not follow from the material what part of the annual employment decline is caused by AI specifically, rather than by restructurings or the general state of the market. It is also unknown how often automatic payout calculations turn out to be erroneous and how successfully clients challenge algorithmic decisions.

Sources

Author

Look at AI, editorial team