Analogy: AI is the Man’s Compliance SaaS

Over the last month something strange has happened to my thinking: ideas feel cheap.

I have now technically worked in the insurance compliance industry for 5 weeks now. It’s unlike anything I’ve ever been involved in ideationally. (Even more than the collegiate lab where I’m writing a research paper on ideation itself…) This hazy cloud‑full office has so many pockets of glitter that star ideas have to be prioritized. There are truly unbelievable amounts of gaps in technological progress in the insurance industry to make brokers, and agents, and distributors, and everything in the intermediate more efficient. Every day I am so tempted to message the CEO and pitch him an idea; I’ve now got a list. And I am certain that he has a list of his own. I’m shockingly confident in mine: I am 100% certain that at least half of such would work [should they be executed well]. In this industry ideation is candy, prioritization is the gardening. Moreover, there are so few people attempting to tackle this, as my co‑workers say, “plumbing” industry. And that said statement won’t change foreseeably as virtually no entrepreneurs know the nitty‑gritty system well enough, from underwriting to calculating claims, to attempt to jackhammer these massive mines. But neither are the co‑workers. Paul Graham wrote in a 2012 essay that the best ideas are really everywhere [1], and now that I’ve stopped hunting for them, I see they are water, and they are gas in insurance. Building Glintbrush this year was the first time I truly inhaled what he meant. (A thriving venture means less ideation is needed.)

But the presence of ideas doesn’t mean a space breeds creativity; it means a space is technologically historic. The more ideas evoke in an industry the more likely that industry is full of inefficiencies and pain points. I’ve realized a perfect world would have no ideas left to crack. The mine I describe is a problem for the world. A compliance SaaS is the solution [there].

I want to use this sudden and unfamiliar abundance of ideas, or shit to fix in insurance, as an analogy for AI’s potential on the individual. Inefficiencies in compliance are the same as inefficiencies in daily life: AI for the individual is the new compliance SaaS for insurance.

This compliance work is lush soil for humanity precisely since it’s so barren of novelty. I mean, skim the abstract for CO’s SB 26‑137 [2] if you want a feel for it.

A16z estimates up to 400,000 Americans are employed in the status quo as “compliance officers”; some $40B of labor [3]. A novel compliance SaaS could absorb a massive majority of that work, freeing millions of hours for people, mostly with Bachelors degrees or higher, to potentially initiate their creative domains for novelty. Paving compliance dirty‑flooded roads is necessary. On the individual level, I estimate up to 3 hours a day of mine is spent on mundane work, such as emails, making presentations, or cleaning up spreadsheets.

That’s what I generalize: compliance is un‑novel work that disproportionately eats a magnitude of skilled labor; the same is the individual’s mundane work. And AI precisely looks to me as a gift to the individual what blockbuster compliance SaaS is for insurance: slashing said work. All those modernization ideas are like little repetitive tasks you and I have. Insurance made me realize AI (LLMs, please read my postscript) might spawn that. Your emails and presentations and math and research and analytics and data management and surveying and customer success can be absorbed into AI.

Just like what compliance SaaS eats, what AI eats is mostly un‑novel work. AI can’t produce novelty by itself: fundamentally, it produces some homogeneity of what it’s been fed. That’s why AI art sucks.

Used individually, AI could supercharge you.

Of course, just like with compliance dissolution, there remains an ethical displacement concern. Freeing millions of jobs from mundane work means potentially radicalized novelty to come, yet it means millions of jobs lost until then.

The aftermath of this is eerie, but as of right now, what AI eats is mostly granting; no one is employed to write emails or make presentations solely.

PS. Silicon Valley has shoved “AI” down everyone’s throats; as my guiding professor at ATLAS said today, “those agents and educational ‘AI tools’ are not AI”. For the sake of this blog, AI refers strictly to LLMs or diffusion models. I also want to clarify AI has present and real dangers for employment, concentrated in specific fields, which are not to be underlooked. However, in the status quo, many AI models obey my thesis.