top of page

AI is Just a Normal Tech

Oct 26
2 min read

Updated: Oct 27

After reading Torsten Slok's Daily Spark post from last week on how AI is not showing any signs of payoff outside of the tech industry (check it out here), I have thoughts.


A chart of total factor productivity vs labor output per hour showing no signs of AI boosting productivity growth.
Sources: Federal Reserve Bank of San Francisco, US Bureau of Labor Statistics, Apollo Chief Economist. Note: Total Factor Productivity (TFP) is the utilization-adjusted series from the San Francisco Fed (Fernald, 2014), adjusted for cyclical variation in factor utilization.

I'm not really surprised by this. I keep hearing the same story from friends and clients in the corporate world: their boss (or boss's boss, or someone up the food chain) gets enamored with the promise of AI. Directives come down to use AI for everything and anything. Emails that were 1-2 sentences are now paragraphs of AI slop.



In some cases, using AI does speed up part of the process. But by virtue of AI being overly verbose, lacking taste or being wrong in ways that are difficult for humans to spot, it makes other parts like reviewing work take a lot longer. And those parts are a lot less interesting for most people (see Cory Doctorow's Reverse Centaur). And that doesn't even get into the productive time lost trying to shoehorn LLMs into workflows that are better handled using old school automation.



In fact, I would bet that the effects of the above are not just a wash but are actually hurting productivity.



It's not all bleak. Ironically, given all the fearmongering about a SAASpocalypse, software that incorporates LLM usage only where it makes sense shows a lot of promise. These products are using LLMs to allow users to interact with the software using natural language instead of navigating UIs, they're still using deterministic calculations when exact answers are needed, and they maintain the security and reliability guardrails that enterprises rely on.



I would never vibe code anything that touches the client experience or personal data, but I have onboarded some advisor tech recently with built-in AI capabilities (Income Lab and SlantAI in particular). I don't really know how much of the features are LLM-driven vs old fashioned automations, but they get the job done. They reduce the time spent on soul sucking tasks and free up time for the parts of my job that are interesting. And that's really all that matters.



The bottom line is this: AI is a normal technology. It's not magic, it's not right for every task, and it's not going to replace humans at scale.



This is bad news for the AI trade. Trillions of dollars have been spent on the promise that this is all going to result in AGI (artificial general intelligence). It's not. No matter how good the magician gets, he'll never actually cut his assistant in half and put her back together for real.



Right now the tech is good enough that, with the right harnesses, it can marginally improve productivity in specific areas. If you don't believe in AGI, what are the frontier models chasing, and what happens when the models powering AI integrations in software are commoditized?



I don't know, but I think we're going to find out soon.

bottom of page