For most of the twentieth century, becoming a skilled insurance underwriter required a decade of apprenticeship. You learned to read a commercial property application the way a doctor reads an X-ray—scanning for the subtle asymmetries that indicated trouble. A warehouse storing lithium batteries next to a wood-frame building. A trucking company with suspiciously low mileage claims. A restaurant owner who had filed three slip-and-fall claims in five years at three different establishments. The knowledge was hard-won, passed down through mentorship, and impossible to fully codify.
That world has not vanished, but it has been fundamentally reorganized. Today's underwriter increasingly resembles an editor reviewing a draft rather than an author writing from scratch. The first pass—sometimes the first several passes—is performed by machine learning models trained on millions of historical policies, claims, and external data sources. The human's job is to approve, adjust, or override.
The shift in daily practice
At major insurers, the transformation began with personal lines—auto and homeowners policies where the variables are relatively standardized and the stakes per policy are modest. A model can ingest a driver's record, credit history, vehicle type, and zip code, then produce a premium recommendation in seconds. Human review became exception-based: flagged applications, edge cases, and appeals.
The more consequential shift is now underway in commercial lines, where policies cover everything from construction projects to cyber liability to directors-and-officers exposure. These risks are messier, the data less uniform, and the judgment calls more consequential. Yet even here, algorithmic triage has become standard. Models score incoming submissions, prioritize the most promising, and pre-populate coverage recommendations. Underwriters who once spent hours building a quote from first principles now spend that time interrogating a machine's reasoning.
What expertise means now
The optimistic view holds that AI handles the drudgery while humans focus on genuine judgment—the complex accounts, the novel risks, the relationships that require nuance. There is truth to this. Senior underwriters report spending more time on the intellectually interesting work and less on routine renewals.
But there is a darker possibility that the industry is only beginning to confront. If junior underwriters no longer learn by doing—if they enter the profession as reviewers rather than practitioners—where does the next generation of expertise come from? The models themselves were trained on decisions made by humans who learned the old way. What happens when that institutional knowledge is no longer being replenished?
Some insurers have recognized the risk and created deliberate training programs that expose new hires to manual underwriting before they graduate to algorithmic oversight. Others have not. The divergence may not matter for years, until the next truly novel risk category emerges and the industry discovers whether its human capital has atrophied.
Our take
The underwriter's transformation is a preview of what awaits many white-collar professions: not replacement, but redefinition. The question is whether we are building a sustainable hybrid model or simply coasting on accumulated human expertise while failing to cultivate its successor. Insurance, with its long feedback loops and catastrophic tail risks, may be the wrong place to find out the hard way.
