It's pretty incredible to see how, in such a short amount of time, AI became part of our lives and seems to be turning into our new go-to monthly subscription. It's also incredible to see how companies keep improving their products to weave those AI subscriptions in so seamlessly that AI seems to be doing everything on its own.
At a conference I attended a few weeks ago, I heard a phrase that stuck with me for a good while: "behind every prompt, there's a human being." So in this article, I want us to think through that statement together.
The giant-brain myth
Many of us believe AI is some giant machine with a brain the size of half of the United States, guzzling water like crazy and cranking out requests for 80s-style pictures or Studio Ghibli art. And while there's some truth buried in there, the reality is that AI doesn't "stick around" all the time.
Our finite brains assume AI knows today is today because the calendar says so, but that's not how it works. AI functions a lot like starting a session in a video game: as long as the game sits in your library, it doesn't "live," but the moment you launch it, it comes to life and "lives" for as long as you're playing. When you say "hey, make me a Ghibli-style image with this reference," it spins up, handles your request, and stays listening. When you're done and delete the conversation (or simply stop talking to it), it shuts down and goes back to being code — the library where your games sit unplayed.
A time machine, but only when you ask for one
AI is an incredible tool. It's the closest thing we have to time travel: it lives in the exact moment you ask it to. If you ask it about something that happened 100 years ago, its memory is as fresh as if it were living through that moment right now. But do you see the pattern? We always say AI "does" something whenever we tell it to. Never the other way around.
And that might be the most important part of this whole reflection: AI doesn't decide when to wake up. It has no curiosity of its own, no to-do list it wants to tackle on its own initiative. Every time it "exists," it's because a specific human, at a specific moment, decided it should.
There are more humans behind every prompt than you think
But that phrase from the conference goes even further than just "you wrote the prompt." Behind every response an AI gives you, there are layers and layers of human work we almost never see:
- Someone wrote the text it was trained on. Every book, article, forum post, and line of code AI "learned" from was written by a person, at some point, for some reason.
- Someone decided which responses were good and which weren't. Today's models go through a process called RLHF (reinforcement learning from human feedback), where real people rate thousands of responses to teach the model what tone, accuracy, and behavior is expected of it.
- Someone decided what question to ask, and how. The prompt doesn't just appear out of thin air: there's a human need behind it, a real problem someone's trying to solve.
So when you ask AI for help with something and it "seems" like pure magic on its own, what you're really seeing is the tip of an iceberg built almost entirely by people.
The physical cost almost nobody sees
And this is where it's worth grounding the "giant brain guzzling water like crazy" part — because, even though it sounds like an exaggeration, there's a real basis to it, just far more nuanced than it seems.
The per-query water consumption figure is, in fact, one of the most disputed numbers in the industry right now. A 2023 study out of UC Riverside estimated that generating a 100-word response could consume the equivalent of a bottle of water (around 500 ml), a figure that went viral. But that number reflected the models and data centers of that era. OpenAI, for its part, claims that today's average query consumes just 0.3 ml of water — a tiny fraction — though that figure only counts direct data-center cooling, not the water used to generate the electricity powering it. Depending on what's being counted and which model is used, current estimates range from a fraction of a milliliter to several milliliters per query.
The honest truth is that the exact number remains genuinely disputed among researchers, and any figure you see floating around online is probably serving some narrative (whether "AI is an environmental disaster" or "AI is basically free to run"). What's undeniable is that, at a global scale — billions of daily queries — even the smallest number becomes significant, and behind that consumption there are human decisions too: where to build the data center, what kind of cooling to use, how efficient to make the model.
AI isn't magic, it's accumulated human work
At the end of the day, the idea that "behind every prompt, there's a human being" isn't just a nice conference soundbite. It's a reminder that, no matter how much AI seems to act on its own, every one of its capabilities was put there, trained, corrected, and constrained by people. AI might eventually "live" in every corner of our home and know what to do without being explicitly asked, but even that future will have been designed, trained, and decided by a human behind the scenes.
And speaking of resources and how much every prompt costs the planet, in our next article we're going to talk about how to optimize token usage in your projects with Graphify — a tool that turns your code into a queryable graph so AI doesn't have to reread your entire project every time you ask it something. And yes, even there, we'll find more humans behind the scenes.
Have a great week, and the next time AI surprises you with a brilliant answer, remember that somewhere, someone had to teach it how to be that good.