Just say the word

Last night, as one does, I was looking at the McMaster-Carr website. A work of beauty, that website. Flat, unhierarchical. Anyway, sitting there, I was wondering — why haven't I seen more software that feels like that?

So I did the most logical thing: I asked Claude.

Just say the word.

At the moment, I didn't know why. But the follow-up question made me feel uneasy.

Weird. I mean, it didn't do anything wrong, really; this machine offered what seemed to be a helpful response (US Graphics is a great website, I've been a fan for a long time!). But somehow, while being seemingly helpful, this automated initiative felt intrusive and, frankly, insulting.

The more I thought about it, the more I realized that this was not the first time I had felt this way. Computers very often behave in ways that you really wouldn't expect another human, or even your tools, to behave.

Tools augment your capabilities

Tools, you see, are items you use to augment your capabilities. You set a specific goal — say, you want to drive a nail into a piece of wood — and you use a hammer to extend the energy from your hand.

Some tools are passive and some are active. A hammer is idle until you pick it up; a dishwasher runs on its own once you start it. But even the active ones only act inside the job you gave them, and you retain full control of the narrative. Because the goal is never just to get the nail into the wall — that's a subtask of your grander goal of renovating a room, which in turn might be a subtask of something else.

The same is true of intellectual work. You have something you want to figure out, you construct your own path to it, and the tool — app, hammer, chatbot — exists to move you along that path.

What starts to happen when the hammer suggests a wall?

All trees, no forest

You lose track of the bigger picture. As long as your AI doesn't have context of the parent goal of what you're working on, its suggestions feel attractive and like the next logical step — after all, prediction is what LLMs are good at — but they're all trees, no forest. Instead of the AI predicting what I will say next, which, fine, whatever, it is prompting you! As if YOU are the token predictor!

Let’s take another, non-AI, example. A few years ago, I purchased a printer¹ on Amazon. A week later, on their home page, my entire page was filled with printers. No, Jeff, I didn’t start a printer collection. I just needed the one.

But I want to take the maximalist position. Even if Amazon recommended toner instead of another printer, even if AI has absolute perfect context and knows what you're trying to do, having something else drive your story is anti-social.

When a tool assumes you are lost and presumes to know your needs, it stops treating you as a focused individual with clear intent and instead treats you as a passive slop consumer waiting to be spoon-fed random directions.

Cognitive offloading

I've recently read about cognitive offloading, which is pretty much what it sounds like but with a decently fancy word. When people start relying on external aids like GPS apps* (about which I've written in the past, it's still on my LinkedIn and I hope to bring it over soon here!) it usually weakens our internal capabilities for spatial reasoning for examples.

I want my machines to know that they're machines, or at least not pretend otherwise. I'm dreaming of tools that wait for my command and act as silent, capable partners, not as eager, presumptuous directors.

Just say the word?

I will. That was always the deal.