Somewhere in the country this week, a plaintiff’s attorney is reading a claim file with a highlighter, looking for the sentence that will cost a carrier six figures. Call him Billboard Bob. You’ve seen him before: Big smile, bigger ads and a habit of turning small details into very expensive problems. Bob isn’t reading the whole file. He’s looking for one paragraph, the moment an adjuster relied on output from an AI tool without documenting what it was, how it works or who validated it. He’ll find it. Most files now contain it.
What Bob does with that paragraph, in written discovery, in a corporate representative deposition, in the first 20 minutes of mediation, is going to cost carriers, and not a little. Many adjusters haven’t read the documentation on the tools in their workflow. Many couldn’t name the models if asked. That gap is about to become a much bigger problem, much faster, than the industry is ready for.
This isn’t hypothetical. On the expert witness side of the table, the pattern is already visible on more bodily injury files than not, and the industry is only beginning to see it. This is the underreported truth of AI in insurance in 2026: 18 months into the most aggressive adoption wave the industry has ever seen, the adjuster’s job didn’t disappear. It got rearranged.
AI doesn’t eliminate jobs. It eliminates tasks. The adjuster role is a collection of tasks, some repetitive, some analytical, some deeply human. Those tasks are being pulled apart, and what’s left is the part that actually decides outcomes.
What the Vendor Slides Got Right (and What They Missed)
To be fair, the pitch to claims leadership wasn’t wrong. AI did make things faster. Intake is cleaner. Summaries are quicker. Fraud signals are seen earlier. Cycle times are compressing in ways that will show up on the combined ratio. That part worked.
What didn’t happen is the part that lived mostly in vendor slides: The vanishing adjuster. Algorithms are very good at the 80% of a file that looks like every other file, and consistently unreliable on the 20% that actually matters. That 20% is where the job lives. It always has.
What Is Quietly Automated Now
The administrative layer is going first, exactly as expected. First notice of loss. First-pass medical summaries. Coverage checks. Duplicate claim detection. Low-severity property damage review. Routine correspondence. Diary notes. At this point, this isn’t innovation; it’s table stakes. An adjuster still summarizing a 400-page medical file by hand isn’t working without AI. They’re working against it.
None of this is the interesting story. These tasks were always structured and repeatable, and also how a large portion of the day got filled. As that layer disappears, the day collapses toward the hardest parts of the job.
The Job Isn’t Lighter. It’s Denser.
A file that once took hours to build now shows up mostly done. Timeline structured. Documents surfaced. Key facts summarized. Sometimes even inconsistencies flagged. So, what’s left? Not building the file. Interpreting it.
That changes the job in a specific way. The question is no longer “What happened?” It’s “Does this actually make sense?” Do the injuries line up with the event? Does the sequence hold together under scrutiny? Is the exposure justified? Not narratively. Mechanically. Medically. Financially.
That’s not process work; that’s judgment. And that’s where the job is concentrating. Automation didn’t simplify the job. It concentrated the risk.
Where the Work Is Getting Harder (and Riskier)
The real shift isn’t automation. It’s scrutiny. High-severity bodily injury files now regularly include analysis generated, at least in part, by algorithms. Causation. Mechanism. Valuation models built on comparables. Meanwhile, Billboard Bob has adapted faster than most carriers expected. Discovery now asks about the tools. Corporate representatives get probed on them in deposition. Mediation conversations increasingly open with them. And in bad-faith litigation, the claim file becomes the case.
That changes the job in a fundamental way. Adjusters are being asked to build files that defend reasoning they didn’t personally generate. That’s new, and it’s uncomfortable, because it’s not a process skill. It’s closer to what an expert witness does: Explain the methodology, acknowledge the limitations and make a call on where it applies. AI doesn’t remove judgment. It puts it on display.
What Is Not Changing at All
For all the focus on automation, parts of the job remain untouched. Sitting across from a claimant and delivering a denial. Holding the line when emotions take over. Reading a room in mediation when a number gets floated and deciding whether the reaction is real, or theater. No model does this, not because it’s impossible, but because it’s human in a way that doesn’t translate into inputs and outputs.
These moments were always central to the job. Now they matter even more. The job didn’t get lighter. It got more human.
The Skill No One Is Training For
A capability is quietly becoming essential, and almost no one is formally teaching it: how to interrogate AI. Not use it. Question it. What does this output actually mean? When is it wrong? When should it be overridden? And how does an adjuster document that override so it survives discovery?
The adjuster who can say “the system flagged this at 87% soft tissue probability; I reviewed imaging and disagreed for these reasons” is building files that hold up. The one who can’t is building files that don’t. Information is no longer the bottleneck. Judgment is.
The Missing Rung No One Wants to Talk About
There’s a structural problem forming, and it isn’t getting much attention. The work AI is replacing is the same work junior adjusters used to learn on. Low-severity files. Summaries. Repetition. That wasn’t busywork. That was training.
Ten years ago, a first-year adjuster would close 40 soft-tissue auto claims before touching a complex bodily injury file. They weren’t learning procedure. They were learning normal, building the mental baseline that lets an experienced adjuster look at a claim and feel that something is off before they can articulate why. Today those 40 files never hit their desk. The new adjuster’s first real case is often their first hard case. They’re being asked to spot the abnormal without ever having seen normal.
The result is a bench that’s getting top-heavy. Senior adjusters with real judgment are more valuable than ever. The pipeline behind them is thin, and the adjusters in it have fewer reps than any generation before them. Automation didn’t eliminate the role. It removed the path into it.
What Adjusters Can Do in the Next 90 Days
This isn’t theoretical. It’s already happening. The response isn’t to “learn AI.” It’s to adjust how the work gets done.
Start with the tools in your workflow. Not at a surface level; actually understand how they work and where they fail. Treat outputs as inputs, not conclusions. Document decisions with the assumption that someone like Billboard Bob will read them later, because he will. Sit down with claims counsel and ask, concretely, what happens when an algorithmic output becomes the issue in a real file. Get the answer before it matters. And identify one junior adjuster on the team and teach them the parts of the job the tools can’t do, because nobody else is going to.
Then, before closing any file with AI-assisted analysis, ask one question. Could you explain your reasoning to a jury at a kitchen table? Not the science. The reasoning. If you can’t, rework the file until you can. That’s the job now.
What This Really Means
AI isn’t replacing adjusters. It’s separating them. The routine work is disappearing. The difficult work is becoming the job. The skill is shifting from processing information to making decisions under uncertainty, with better tools and much less margin for error. From the outside, the role will look the same. From the inside, it won’t.
Because the job isn’t going away. The margin for getting it wrong just disappeared.
Rami Hashish is a forensic biomechanist and CEO of Silent Witness. He has been retained on roughly 1,000 personal injury and product liability matters and sits on the Association for the Advancement of Automotive Medicine’s Abbreviated Injury Scale committee.
