The Adjuster That Never Sleeps

For most of the past century, the legal spine of an insurance claim was made of paper and patience. A first notice of loss arrived, a human adjuster opened a file, a coverage lawyer read the policy clause by clause, correspondence shuttled back and forth, and somewhere in that slow choreography a decision emerged, pay, deny, or fight. That process produced the modern doctrine of bad faith: the legal duty an insurer owes to handle a claim reasonably, promptly and honestly. Now a new kind of worker is entering that choreography, not a chatbot that answers questions, but an autonomous agent that takes actions. These trigger-based systems read the policy, gather the evidence, draft the position, and escalate only when they hit the edge of their own confidence, logging every step as they go.

The shift is no longer hypothetical. Industry analysts forecast that the enterprise market for agentic software will climb from roughly $1.5 billion in 2025 to $41.8 billion by 2030, a curve steeper than generative AI's own early ascent, according to Omdia. Insurance, a business defined by documents, rules and disputes, sits squarely in its path.

Sources: NAIC; Gartner via Reuters; McKinsey; Coalition Against Insurance Fraud.

The Old Way: Files, Folders and the Long Wait

Legacy claims-and-coverage work was built for a world of scarcity, scarce data, scarce specialist time, scarce ways to coordinate them. A complex liability claim could take weeks simply to route to the right reviewer, and assessing liability on a difficult file often consumed three or four weeks of back-and-forth before anyone could quote a reserve. The bottleneck was not bad intent; it was the sheer friction of moving unstructured information, medical records, demand packages, adjuster notes, engineering reports, past human eyes one document at a time.

That friction had legal consequences. Insurance law in most U.S. states recognizes a tort of bad faith precisely because delay and inattention are themselves a kind of harm. When an insurer "unreasonably delays or denies payment," many states impose liability regardless of whether the conduct was willful, and a policyholder may recover not only the benefit owed but consequential and even punitive damages, as legal analysts summarizing the doctrine note. Manual handling created the very conditions, missed deadlines, ignored evidence, inconsistent decisions, that the doctrine exists to punish.

Detection of the costliest problem, fraud, was equally slow. Fraud across all lines is estimated to drain $308.6 billion a year from the U.S. economy by the Coalition Against Insurance Fraud, with property-and-casualty fraud alone near $45 billion; the FBI's narrower estimate of roughly $40 billion annually still ranks insurance fraud among the largest economic crimes in the country. Schemes typically ran some eighteen months before detection, a lag that human-paced review could rarely close.

The Shift: From Copilot to Colleague

The defining feature of an agentic workflow is not that it answers, it is that it acts, in sequence, toward a goal. Where a conventional tool performs a single function on command, an agent chains analysis, assessment, decision preparation and document drafting into one process, as practitioners describe the category. In a claims context that means an intake agent ingesting a loss notice, a coverage agent mapping facts to policy language, and an orchestrator deciding whether a file can be settled automatically or must escalate to a senior human reviewer, a pattern consulting researchers describe as a multi-agent "decision orchestrator," per McKinsey.

Adoption data shows how quickly the ground is moving. In NAIC survey results, 88% of responding auto insurers and 70% of home insurers reported using, planning to use, or exploring AI/ML models, with health insurers at 92%. Across the broader professional landscape, Gartner projects that at least 15% of day-to-day work decisions will be made autonomously by agentic AI by 2028, up from effectively zero in 2024, though the same firm warns that over 40% of agentic projects may be scrapped by 2027 where they solve the wrong problem.

The legal profession that serves these carriers is moving in parallel. Active generative-AI use among law firms rose from 14% in 2024 to 26% in 2025 in Thomson Reuters Institute survey data, with corporate legal departments at 23%, and document review, long the most labor-intensive task in litigation, has historically consumed roughly 73% of e-discovery production costs. That is exactly the high-volume, bounded work agents are built to absorb.

An agent that can read a policy, gather the proof and draft the denial in minutes does not abolish the duty of good faith, it industrializes the moment at which that duty is tested.

What It Looks Like Now

Strip away the marketing and a present-day agentic claims workflow is a disciplined relay. A trigger, a new loss, a litigation hold, a regulatory deadline, wakes the system. Specialized agents then move through defined steps within hard guardrails: extract the facts, check them against coverage terms and jurisdictional rules, calculate a position, and draft the correspondence or filing. Crucially, the workflow is designed to stop. When confidence falls below a threshold, when a clause is ambiguous, or when the dollar value crosses a set limit, the file escalates to a human, and every input, model version and reviewer identity is written to an immutable log.

That audit trail is not a nicety; it is increasingly a legal requirement. The EU's AI Act, whose high-risk obligations take effect for relevant systems by August 2, 2026, classifies AI used for risk assessment and pricing in life and health insurance as high-risk and demands automatic logging of inputs, outputs, model versions and the human reviewers involved in verification. In the United States, the NAIC Model Bulletin on the Use of AI Systems, adopted in 2023 and approved by NAIC members in 2024, expects every insurer to maintain a written program ensuring AI-supported decisions are not "inaccurate, arbitrary, capricious, or unfairly discriminatory."

The efficiency case is no longer speculative. Carriers deploying AI claims systems report processing-time reductions on the order of 60% in research compiled from McKinsey and industry benchmarks, with straight-through-processing rates for simple claims climbing from the old 10 to 15% band toward 70 to 90%. One large carrier's rewired claims journey cut the time to assess liability on complex cases by 23 days and improved routing accuracy by 30%, McKinsey reports. On the fraud side, Bain estimates that full-scale generative-AI adoption could cut loss-adjustment costs by 20 to 25% and leakage by up to 50%.

The Next Few Years: Autonomy Meets Accountability

The trajectory points toward agents that handle whole classes of routine matters end to end. McKinsey's claims research projects that by 2030 carriers will determine liability and appraise 90% of simple claims from claimant-submitted photos and sensor data without a physical inspection, and that straight-through processing for predictable personal-lines claims will exceed 90%. The enterprise agentic market's projected 175% five-year growth rate, per Omdia, suggests the tooling will be there to meet that ambition.

But autonomy and accountability pull against each other, and that tension is where the legal risk concentrates. The bad-faith doctrine does not disappear when a machine makes the call, it sharpens. An agent that systematically applies a "delay, deny, defend" pattern would not mitigate liability; it would manufacture a discoverable, reproducible record of it. The same audit trail that proves diligence can, in the wrong configuration, prove the opposite. And because contract and agency law treats AI systems as lacking legal personality, their actions are attributed back to the humans and corporations that deploy them, as legal commentators on autonomous workflows emphasize.

Regulators are already drawing the perimeter. Colorado became the first state to require deployers of high-risk AI systems to exercise reasonable care against algorithmic discrimination, with its broader AI law and a private-passenger-auto and health governance regime phasing in through late 2025 and 2026. The EU AI Act backs its requirements with penalties reaching up to 7% of global annual turnover for prohibited uses and 3% for high-risk violations. As of March 2026, twelve U.S. states were piloting the NAIC's AI Systems Evaluation Tool, a market-conduct instrument that probes governance, high-risk models and data inputs, per NAIC.

The practical near-term picture, then, is not full autonomy but a widening band of supervised autonomy. Agents will own more of the routine, the clean claim, the standard filing, the first-pass coverage opinion, while humans concentrate on the contested, the ambiguous and the high-value. The deciding variable will be the quality of the guardrails: the escalation thresholds, the bias testing, and above all the audit trail that lets a regulator, a court or an opposing counsel reconstruct exactly why the machine did what it did.

Conclusion

The arc from the paper file to the autonomous agent is, at bottom, a story about where human judgment belongs. The old way spread thin attention across every claim and let the hard ones fester; the new way promises to clear the routine at machine speed and reserve scarce expertise for the files where coverage, fairness and good faith are genuinely in play. That promise is real, and the adoption data shows carriers chasing it hard. But the law is unambiguous that a faster decision is not automatically a better one. The duty of good faith, the prohibition on unfair discrimination, and the demand for an explainable record all survive the transition, and may matter more, not less, when the adjuster never sleeps. The institutions that thrive will be the ones that treat the audit trail not as overhead but as their best evidence that the machine, and the humans behind it, acted reasonably.

Sources

1. McKinsey & Company, The future of AI in the insurance industry
2. McKinsey & Company, Rewiring the insurance claims journey with AI
3. NAIC, Insurance Topics: Artificial Intelligence (2026)
4. NAIC, Model Bulletin: Use of Artificial Intelligence Systems by Insurers (2024)
5. Reuters, Over 40% of agentic AI projects will be scrapped by 2027, Gartner says
6. Omdia, Enterprise Agentic AI Software Market Forecast (2025)
7. Thomson Reuters Institute, 2025 GenAI in Professional Services Report
8. PlatinumIDS, The Rise of Agentic AI in Legal Technology (e-discovery cost data)
9. AI Claims Processing Automation Statistics 2026 (compiling McKinsey & industry benchmarks)
10. IJETCSIT, Generative AI in P&C: Transforming Claims and Customer Service (Bain estimates)
11. InsuranceNewsNet, The rising tide of insurance fraud: an estimated $308B problem
12. Coalition Against Insurance Fraud, Insurance Fraud Costs the U.S. $308.6 Billion Annually
13. GLACIS, Is Insurance Underwriting AI High-Risk Under the EU AI Act?
14. Munich Re, New EU Act Regulates AI in Insurance
15. Faegre Drinker, Colorado Division of Insurance Expands AI Governance Obligations
16. The Legal Wire, The Double-Edged Sword of Agentic AI (legal personality / attribution)