Time and Tokens

Almost everyone is fascinated by the pace of AI. We haven’t finished learning one topic before a leading researcher from a leading lab ships a breakthrough that makes it obsolete. So the natural question is, can we keep up?

Before answering, we need to be honest about what we are looking at. AI grows right under our nose. We can feel it, we can compare it year over year, and the progress is massive. That is a strange thing to live through. But the more useful question isn’t how fast the frontier moves. It is how many people actually use AI at that pace, and how many still treat it as nothing more than a search engine replacement.

AI is a tool. Someone building a nuclear bomb doesn’t mean you need one to demolish an old bridge. The good news is that yesterday’s frontier becomes economically reachable very quickly. The model that was expensive and exclusive a year ago is cheap and boring today. That is the part most people leave on the table. The real gap is not between us and the frontier, it is between what we already have access to and how little of it we actually use.

It helps to split AI into two areas, AI as a science and AI as engineering. Someone builds a great agent, that’s cool. But for most of us the problem statement is not who is creating the leading model. It is how to use some model, economically, to serve our purpose.

AI also highlights a bias worth naming. It can do things, it can get things done, and how impressive that looks depends on your own expertise. The less expert you are in a domain, the more incredible the output seems. Expertise is what tells you whether the work is actually good, or just fast.

The trade off is time and tokens. Ignore the inner architecture for a moment, and those two are the major variables for delivering the same result. You can get something done in an hour while burning ten times more tokens. Is that what we need? Sometimes yes, sometimes clearly not, and the difference between those two cases is worth thinking about before you start.

So let’s ignore the noise about the newest method or approach. The goal is simpler, use AI effectively and efficiently, with economic value as the measure.