Claims adjusters are not Luddites. They are not afraid of technology. They are drowning in repetitive tasks, and they know a tool that adds more clicks, more screens, and more second-guessing is not a tool, it's a tax. The recent coverage of adjuster resistance to AI misses the point. The backlash is not about AI itself. It is about how AI has been deployed: as a black box that makes decisions and then demands the adjuster clean up the mess.
I have seen this pattern play out across the industry. A carrier rolls out an AI model that flags claims for fraud or estimates damage. The adjuster gets a notification that says, in effect, "trust this number." But the adjuster knows the property, the policy, the local building codes, and the homeowner's story. The AI does not. So the adjuster has to open the AI's output, cross-check it against their own field notes, and then override it half the time. That is not automation. That is an extra step.
The Real Source of Friction
The Wired piece on adjusters hating AI captures the frustration, but it misdiagnoses the cause. Adjusters do not hate AI because it threatens their jobs. They hate it because it adds cognitive load. Every AI suggestion becomes another thing to verify, another potential error to catch, another reason to doubt their own judgment. The tool was supposed to save time, but it ends up costing time because it is not designed to fit into the adjuster's actual workflow.
Consider the physical world AI agents that are starting to appear in insurance. These agents need location data, imagery, and property context to make decisions. That is promising. But if the agent's output arrives as a separate report that the adjuster has to interpret and then manually enter into the claims system, you have just created a new data entry job. The agent should be doing the data entry. The adjuster should be doing the judgment.
The Second-Order Problem: Who Is Accountable?
There is another layer here that most operators have not connected yet. When an AI agent makes a mistake, who pays? The emerging literature on AI agent liability points to a gap in coverage. If an autonomous system causes damage, the traditional insurance policy may not respond. That is a legal and actuarial problem, but it is also an operational one. If the adjuster is the one who ultimately signs off on the AI's decision, they are carrying the liability. No wonder they push back.
The solution is not to make AI more accurate. It is to make the human-AI handoff explicit and auditable. The adjuster needs to know what the AI did, why it did it, and what their own responsibility is. That requires a system that is transparent, not a black box. It requires an interface that shows the reasoning, not just the conclusion. And it requires a workflow where the adjuster's role is clear: they are the final decision maker, and the AI is the assistant that prepares the evidence.
What Operators Should Do Differently
If you are running a claims operation, the lesson is not to abandon AI. It is to redesign the workflow around the adjuster. Start by asking what the adjuster actually does all day. Then ask which of those tasks can be automated without adding friction. The answer is usually not the big judgment calls. It is the small, repetitive steps: pulling policy documents, extracting data from PDFs, checking for missing signatures, drafting initial correspondence.
Those are the tasks where AI agents shine. They do not need to make the final call. They need to prepare the file so the adjuster can make the call faster and with better information. That is the difference between an AI that replaces the adjuster and an AI that supports them.
I have seen this work in practice. When you give an adjuster an agent that pre-fills the estimate template, attaches the right photos, and flags the policy exclusions, they do not resist it. They ask for more. The resistance comes when the AI tries to do the whole job and then hands the adjuster a mess to clean up.
The Opportunity for the Industry
The insurance industry is at a turning point. The technology is finally good enough to handle real-world complexity. But the adoption problem is not technical. It is human. The carriers that figure out how to design AI agents that fit into the adjuster's workflow, rather than around it, will see the productivity gains they are hoping for. The ones that keep deploying black boxes will keep getting pushback.
The same principle applies to the liability gap. If you build an agent that acts autonomously, you need to define the boundaries of its authority. You need to know who is responsible when it makes a mistake. That is not just a legal question. It is a design question. The agent should be built to fail safely, to escalate to a human when it is uncertain, and to leave a clear audit trail.
A Better Way Forward
At SketchMyRoof, we have spent years building measurement tools that adjusters actually use. The lesson we have learned is that the tool has to fit the workflow, not the other way around. That is why Built by SMR exists. We build AI agents and custom portals that are designed for the people who use them, not for the technology.
This is the kind of workflow design problem Built by SMR is being asked to solve for carriers and restoration companies right now. We build agents that prepare the file, not replace the adjuster. We build portals that show the reasoning, not just the result. And we build integrations that put the AI output exactly where the adjuster is already working. That is how you get adoption. That is how you get the productivity gain without the backlash.