Essay · · 3 min
You Had to Be There
Everyone calls it AI, but what we mean depends on what we've seen it do.
I was talking to a friend one evening. I was fresh off a day at work where I played commander with Claude. I was nearly giddy with how I had created a “chief of staff” Claude agent to manage my to-dos. As we talked, I became disoriented: I hadn’t gotten to unpack the details of how I had “hired” my chief of staff before my friend was talking about a laissez-faire government approach to data centers. How did we get so off track?
Turns out, we were both talking about AI.
When I took graduate courses at Georgia Tech, one of my professors told us a story about AI. When MapQuest was everywhere around 2000, it revolutionized how we got around town. Instead of tracing routes on fold-up maps, we printed off turn-by-turn directions calculated by MapQuest, complete with exact mileages. It was such a leap forward that people at the time called it AI. Fast forward a few years and we had grown accustomed to MapQuest, Google Maps, Garmins. We stopped calling it AI. When IBM Watson won Jeopardy! in 2011, it too was endowed with the AI moniker. By the mid-2010s, Netflix’s “Because you watched…” had become a staple in our living rooms, and prediction algorithms became too ordinary to be AI anymore.
“AI” is not a destination. It’s simply a mark in the sand representing what wows us. Once the luster fades from a new technology, the mark is erased and redrawn further out.
But you can’t be amazed by what you haven’t seen. This is a technology people are experiencing in different doses. Some people are drinking watered-down drip coffee, others are throwing back double espressos, and some have sworn off caffeine altogether.
For most people, LLMs haven’t earned the AI label yet. Someone who only reads about AI hasn’t seen what it can do. Neither has someone who tried ChatGPT in its first few months, nearly four years ago. They only see the endless stories about data center buildouts and concerns about a bubble. Those who’ve used LLM chat agents more recently recognize them as an upgrade over Google search. The real magic, where LLM agents convincingly claim the AI label, is in the hands of practitioners: people who use them day in and day out. They appear as wizards orchestrating LLMs equipped with tools to write code or dig through reams of text for the needle in a haystack.
In my conversation with my friend, we couldn’t have been discussing the same thing. AI companies and their investors are selling the idea that LLMs are AI and the pitch is like sowing amongst the rocks. It withers practically before it sprouts. Without seeing the payoff, people assume the spending on data centers is wasteful at best.
It’s nearly impossible to describe a magic trick to someone who wasn’t there. Without seeing it firsthand, should anyone be surprised it looks hollow? And yet, someday soon, my agentic chief of staff will be as quaint as MapQuest is today, while the mark in the sand is redrawn somewhere new.