The Ghost in the Machine: Reclaiming Human Agency

When I first used a large language model to help me draft an email, I felt an unsettling mix of relief and unease. Relief because it worked — the output was good, sometimes better than what I’d have written. Unease because I couldn’t quite locate where I ended and the machine began.

This is the central question of our moment: not whether AI is intelligent, but what it means for us if it is.

The Automation of Intuition

For most of human history, the things machines could do were the things we understood mechanistically. A loom weaves thread because we can describe exactly how the shuttle moves. A calculator multiplies because we can specify the algorithm to the last bit.

But language models don’t work like that. They produce outputs we couldn’t have predicted, make connections we wouldn’t have made, and sometimes feel — there’s no better word for it — insightful. We’ve built systems that simulate the most distinctly human parts of thinking: creativity, analogy, narrative.

This is new. And it changes things.

What Gets Automated First

There’s a pattern to how automation spreads. It starts with the most repetitive tasks, then moves up the value chain. The question isn’t whether a task will be automated — it’s when, and what we do with the freed-up capacity.

With language models, we’re not automating the mechanical parts of writing. We’re automating the parts that felt irreducibly human: synthesis, judgment, voice. Not the typing. The thinking.

Reclaiming the Spark

I’ve come to believe the answer isn’t to resist these tools, but to use them in a way that keeps our judgment upstream. The model can draft; you decide what’s worth drafting. The model can generate; you curate what matters.

The skill that doesn’t automate away is knowing what you want to say — having a perspective worth expressing. That’s harder than it sounds. Most of what we write is just social performance: saying what’s expected, in the expected way. Strip that away and you’re left with the raw question: what do I actually think?

The best use of AI, I suspect, is as a mirror. Write with it, then read what comes back and ask: is this what I meant? If yes, maybe you didn’t have much to say. If no, the gap between output and intention is where your real thinking lives.

That gap is where agency lives.