Working with AI: tool or collaborator?

One of my channels is largely devoted to AI, and I work with these tools every day. Despite that, or perhaps precisely because of it, I keep asking myself the same question: is this thing my tool, or my collaborator? My answer has shifted over time, and this post is its current form.

Where the speed genuinely helps

There are tasks where AI is unarguably good: first drafts, repetitive boilerplate, seeing ten variations of an idea, sketching a rough map of a field I know nothing about. The first eighty percent of the work accelerates dramatically. The "blank page" stage that used to cost half a day now takes minutes. And the blank page was always the most expensive stage of making anything.

Taste is the new bottleneck

But a balance has shifted here: as production got cheaper, choosing got more expensive. AI gives you ten options; it doesn't tell you which one is good. Knowing what's good, what the industry calls taste, judgment or vision, is now the real bottleneck. And taste is a muscle with no shortcut: it grows over years, by studying great work and by making bad work and feeling embarrassed about it.

Don't outsource the thinking

The most real risk I see is not the work being delegated, but the thinking. Getting an answer and understanding an answer are different things; every answer used without understanding is a debt that gets repaid with interest. My personal rule is simple: AI produces drafts, not decisions. The final sentence I must always construct with understanding, because the signature is mine.

Tool or collaborator? My current answer: neither; a powerful apprentice. It accelerates the work extraordinarily; but knowing what's worth making remains, and I suspect will always remain, the master's job.