AI and the Modern Agent: Tool, Threat, or Transformation
Three months ago I watched a broker close a quote in eleven minutes that would have taken her team eight hours. The next morning, in the same kind of brokerage, a different broker was on the phone with a client who had run his own AI-generated benefits analysis and was now arguing — convincingly — that he did not really need a broker at all.
Both things happened. Both things will keep happening. The question is what an agent is supposed to do about it.
There are three frames in circulation right now for what AI means for the modern insurance agent. The frames are usually treated as competing — pick one — but I think they are all true at the same time. Choosing one and ignoring the other two is the most common mistake I see being made, by agents and by the people writing for agents. So let’s hold all three.
Tool
AI as tool is the safest frame, and the one most likely to be true in the next twelve months for most brokers most of the time. The work an agent does in a given day — summarizing a renewal, building a comparison grid, drafting a follow-up email, normalizing a census, pulling rates, flagging an anomaly in a claims report — is roughly seventy percent the kind of work that current-generation AI is reasonably good at, and roughly thirty percent the kind it is reasonably bad at.
The seventy percent is real. I have seen hours-to-minutes step changes in specific tasks. Document parsing of an SBC, when prompted well, lands around ninety to ninety-five percent accuracy on the structured fields and produces a normalized output a human can verify in a fraction of the time it would take to extract from scratch. Census reconciliation against a renewal — the kind of work where one person used to spend half a Friday — runs in under a minute, with the AI flagging edge cases for human review instead of having a human discover them on Monday. Email drafts that used to take fifteen minutes to write now take three to revise. The compounding effect across a week of these is real and visible to anyone who has actually tried.
The thirty percent is also real, and it is the part that the vendor demos quietly skip. Multi-table extraction across documents with merged cells. Carrier-specific underwriting nuance. Anything that depends on reading what is not on the page — the implicit context an experienced broker carries in their head. Tasks where being eighty percent right is actually worse than not done at all, because the error has to be hunted down later. AI is not yet good at knowing what it does not know, which means it confidently does the wrong thing in ways that are particularly hard to catch.
The agents who get the most out of AI as a tool are the ones who learn the line between the seventy and the thirty for their own book of business. That is craft, not technology. No vendor will teach it to you. You learn it by trying, watching the failures, and remembering them.
Threat
The threat frame is harder to talk about, partly because nobody wants to talk about it, and partly because the threat is less imminent than the headline writers say and more imminent than the industry usually admits.
Here is what is true. The pieces of the broker’s value proposition that are easiest to commoditize are the ones AI will commoditize first: comparison shopping, plan summarization, basic recommendation generation, simple compliance checks. A motivated employer with a halfway-decent AI tool can now produce a workable first-pass benefits analysis in an afternoon. It will not be as good as what a competent broker produces. But it will be good enough for some employers, in some situations, to skip the broker entirely.
This is not new. Direct-to-employer carrier sales, online enrollment platforms, and HR consultants have been chipping at the edges of the broker’s role for two decades. AI accelerates the chip. It does not invent it.
The question that is not really a question is this: which parts of your value as a broker depend on doing the things AI is about to be good enough at? If you have an honest answer, you have a plan. If you do not, the next two years will be harder than the last twenty.
The agents most exposed are the ones whose work centers on processing — putting plans in front of clients, summarizing options, doing the work of comparison. The agents least exposed are the ones whose work centers on judgment, negotiation, advocacy, and relationship across the long arc of an account. The shift is not from one to the other overnight. It is from the first being most of the job to the first being a smaller and smaller share of it, year over year.
Transformation
The third frame is the one I think is most accurate and least talked about. AI is not making the agent’s job easier or replacing it. AI is shifting the locus of what the job is.
The shift, in one sentence: the agent’s job becomes less about processing and more about judgment, less about handling work and more about handling the cases where the work is hard.
The transactional middle of the broker’s day — the predictable, comparable, structured stuff — gets compressed. Tools handle more of it. AI handles more of it. The portion of an agent’s day that goes into the predictable middle drops from seventy percent to forty percent to twenty percent, over time, at different rates for different books of business.
What gets bigger is the part that was always the hardest and the highest-margin: the unusual case, the multi-line group, the claims advocacy, the carrier negotiation, the long-cycle renewal where context from three years ago matters and is not in the file. The judgment work. The relationship work.
This is not a comfortable transformation. It demands that brokers be measurably better at the hardest part of the job than they have historically had to be, because the easy part of the job no longer absorbs as much of their time and no longer obscures their performance on the hard part. The agents who already do the hard part well will thrive. The agents who built a career on volume-of-transactions throughput will struggle, because the throughput itself is becoming cheap.
This is not a tide that lifts all boats. It is a tide that lifts the boats already pointed in the right direction. The agents who treat AI seriously now — who learn its edges, who use it to compress the easy work so they can spend more time on the hard work, who build their book around the things AI cannot do — are the agents who will be holding the room in 2030.
Holding all three at once
The mistake is choosing.
Treating AI only as a tool and ignoring the threat is how brokers walk into the next two years unprepared. Treating AI only as a threat and missing the tool is how brokers waste the most productive year of their career. Treating AI only as transformation and skipping practical adoption is how brokers spend a lot of words on the future and almost no time on the work.
The frame that holds up is the compound one. AI is a tool, today, for the work I am already doing. It is a threat, this decade, to the parts of my value that are easiest to commoditize. And it is a transformation, over the longer arc, in what an agent’s job is actually for.
The agents I would bet on hold all three frames at once and act on each of them. They are using AI on their own desk this quarter. They are honestly reassessing their book against the commoditization risk. And they are gradually moving the center of gravity of their book toward the work that gets harder to replace as AI gets better at the work that does not.
I do not know exactly what an agent’s job looks like in 2030. I know it looks less like processing and more like advising. I know the agents who are practicing for that job now will be the ones in the room when the new role becomes the only role.
So the question I keep asking myself, and the one I leave with you: what part of your work is most exposed, what part of your work is most defensible, and what are you doing this month to move from the first toward the second?