Career redefinition

 I had a high prestige job in a downtown SF skyscraper (the B of A building) up high enough that my office overlooked the bridges and Angel Island and incoming ships.  It also meant a nice expense account for taking clients to lunches and dinners and them taking me.  I had a wonderful Victorian "Old Lady" I rented, living there essentially rent-free with a group of eccentric roommates paying the bills.  Very comfortable and a quick Bart commute.

I was respected, active in CPCU, and bored and stressed out.  Underwriting is like a pitcher trying to get the ball over the base without letting the batter hit it.  That and the constant fuss over getting my poet and female member of the local ballet company to pay their share on time had me going, and I decided to redefine myself, as the jargon puts it.

So I sold my employer on sending me to a short course in what then passed as AI and was very much hyped, rule based systems (I already knew a couple coding languages), moved to the company headquarters in Kansas City and became the go-to man for applying rule based systems to underwriting.  Kansas City is not San Francisco.  The winters are bleak and the summers not anything to write home about.  I was, after selling the SF place, able to buy a virtual mansion there, again with roommates, but my office was suburban and no expense account.  The trade was very little stress.

That is quite a vivid example of a career “reinvention” that was both rational and a little strange by today’s standards. You essentially traded status, scenery, and stimulation for control, comfort, and a different kind of intellectual challenge.

The San Francisco part sounds almost like a movie version of late-20th-century professional life: the B of A building, the view over the Bay, clients at restaurants, professional organizations, a Victorian house full of colorful characters. It had a lot of external markers of success. But the interesting thing is that your description of underwriting is almost entirely about the human friction of the job, not the technical side. Your pitcher analogy is a good one: you are trying to throw something that is acceptable to the market, profitable to the company, and not so restrictive that the producer takes his business elsewhere. The ball is the policy; the batter is all the competing pressures.

And then there were the little daily irritations that accumulate. A poet and a ballet dancer sound like wonderful roommates for a novel, but collecting rent from artistic people may be less charming in real life. A lot of stress comes not from the grand problems but from the repeated tiny negotiations.

The move into “AI” is also fascinating because it was AI in the earlier sense: not machine learning, not neural networks, but encoding expertise into rules. In some ways it was very aligned with underwriting, because underwriting itself is often a collection of judgments, exceptions, and heuristics. You were taking something that experienced underwriters carried around in their heads and trying to make it explicit. That was very much the spirit of expert systems in that era.

The Kansas City move also illustrates something that is easy to miss: prestige and quality of life are not the same thing. A downtown San Francisco office overlooking the Bay is objectively glamorous, but a suburban office with fewer demands, a house you actually enjoyed, and the ability to sleep at night can be a better deal. You exchanged being a visible player in a high-pressure arena for becoming a specialist whose value came from knowing something unusual.

There is a slight irony too: you left the glamorous “people” side of underwriting for technology, but your new role still depended on understanding people — the experienced underwriter, the producer, the business rules they used without realizing it. The computer was just the new colleague that never argued over its rent share. 🙂

That actually changes the picture quite a bit. Three school teachers as roommates fits a very different pattern from the stereotypical “bohemian Victorian house” image. It sounds less like a chaotic artist commune and more like a practical arrangement among educated people who enjoyed an interesting shared household.

There is also a nice contrast between the two phases of your life. In San Francisco you had the grand setting — the Victorian, the view from the skyscraper, the dinners, the professional recognition — but your home life was probably grounded by ordinary working people. In Kansas City you had the “virtual mansion,” but the social ecosystem may have been similar: sharing space with people who made the economics work and brought their own personalities into the house.

Teachers are interesting roommates because they often have a strong sense of routine and community. After spending the day managing classrooms, they may actually have been among the more orderly and dependable people to live with. (Though I imagine three teachers together could also produce some very lively discussions about everything from education to politics.)

It also highlights something about your choices: you seem to have repeatedly optimized for interesting environments rather than just conventional markers. A Victorian house with teachers in San Francisco and a large shared house in Kansas City are not the standard executive path of “big house in the suburbs with a spouse and a mortgage,” but they provided something many people never get — a daily mix of different personalities and conversations.

And the work, at least at first, was a lot more fun, if you leave out needing to scrape the ice off one's car windshield every winter morning.

That makes the tradeoff even clearer. It sounds like the Kansas City move was not simply an escape from stress; it was a move into a period where the work itself was more playful.

There is a big difference between being the person who has to say, “No, we can't write this risk unless the premium is higher and the exclusions are tighter,” and being the person who gets to invent a system that says, “What if we could capture how the best underwriters think?” The second one has the pleasure of exploration. You were no longer just throwing the pitch; you were helping design the pitching machine.

And the timing was fortunate. Rule-based AI was one of those moments when the field was full of optimism and possibility. It was small enough that a knowledgeable person could actually understand the whole thing. You weren't one programmer among thousands maintaining some enormous opaque system; you could see the rules, the logic, the business problem, and the people using it.

The windshield scraping is a wonderfully mundane counterpoint to all that. There you were, a San Francisco professional who had looked out over the Bay and watched ships come in, suddenly standing in a Kansas City driveway with an ice scraper. There are few things that make a life transition feel more real than having to negotiate with frozen glass before coffee.

But there is something appealing about the bargain you made. You gave up a dramatic view and expense-account dinners, but gained autonomy and intellectual curiosity. A lot of people spend years chasing the first set of things and only later discover they wanted more of the second. You seem to have recognized that at the right moment.

Being physically in the home office of a company has some advantages over being out in a branch.  In ten years, even though the rule based systems didn't do as promised, they could be used well for well documented normal code.

That is a point that often gets lost in the history of “failed” technologies: a technology can fail at its grand promise and still succeed at changing how work gets done.

The expert systems of that era were often oversold as if they would replace human judgment — the classic “capture the expert’s brain and put it in a computer” vision. That was much harder than people expected. Real underwriting knowledge is full of exceptions, context, intuition, and tacit knowledge that is difficult to express as rules.

But the more modest application was much more practical: take well-understood procedures and make them consistent, documented, and repeatable. A rules engine could be very useful for things like checking completeness, applying standard guidelines, flaging unusual cases, generating explanations, and making sure institutional knowledge did not disappear when an experienced person retired or moved on. That is a much less glamorous promise, but a much more durable one.

And being in the home office would have been a huge advantage. In a branch, you are often focused on your local book of business and immediate pressures. In headquarters, you are closer to the people who design products, set underwriting philosophy, manage systems, and make strategic decisions. You also get exposure to a wider range of problems because everyone knows where to find the person working on the unusual new thing.

There is also a subtle career advantage: being the person who knows something nobody else quite understands gives you a different kind of influence. You are no longer just another underwriter competing with other underwriters on the same measures; you become the translator between technology and the business.

It is interesting that your description mirrors what happened with many early AI efforts. The “AI will replace experts” part faded, but the “AI can help organize expertise and make routine decisions more reliable” part became a foundation of later business software. The hype was wrong about the destination, but not entirely wrong about the direction.