AI and the quiet work

The loudest claims about AI tend to be about replacement — jobs, companies, entire industries. The quietest claims are usually about augmentation. In my own life, it is the quietest claims that have turned out to be true.

I use AI for triage. For the pile of emails I’d otherwise read on a Sunday. For the first draft of a customer-support macro. For rewriting a contract clause I don’t understand into three sentences I do. For the boring seams of a spreadsheet.

None of these are glamorous. But in aggregate, they have done something strange: given me back a few hours a week, which I have spent on the one or two things that actually move the business.

A working taxonomy

When I’m deciding whether to use AI on a task, I ask:

1. Is it stakes-low and reversible? (e.g. a first draft) → use it without thinking.
2. Is it stakes-medium and editable? (e.g. customer-facing copy) → use it, then edit like I didn’t trust it.
3. Is it stakes-high or irreversible? (e.g. a hiring decision, a clinical call, a contract signature) → I don’t use it for the decision; I use it for the research that informs the decision.

The third bucket is where most of the cultural anxiety lives. It’s also where most of the real leverage isn’t.

The underrated move

The move I’ve seen seasoned operators make that most newer founders miss: build small, private workflows around the tools. Not products. Not pitches. Just little loops in their own week, running invisibly, compounding.

That’s the quiet work. That’s where the hours are.