The most expensive part of your AI workflow is the part you didn't automate

Half-automating a task can cost more time than leaving it alone. The reason is quiet, structural, and almost nobody mentions it.

There's a particular disappointment that shows up about three weeks into using a new AI tool.

The tool works. The outputs are fine. And yet the work is somehow taking longer than it did before you had any help at all. You start wondering whether you're using it wrong.

Usually you're not. Something else is happening, and it has a name outside of software: it's the cost of a handoff.

Where the time actually goes

Take a task with five steps. You automate three of them. Those three are now instant — that part is real, and it's the part everyone measures.

But every boundary between automated and manual work creates something that wasn't there before. A moment where you stop, read what the machine made, decide whether it's right, and climb back into your own head to continue.

That climb is the cost. And it never shows up in a before-and-after comparison, because nobody times how long it takes to remember what you were doing.

Which is why partial automation sometimes loses to no automation at all. The manual version holds one thread the whole way through. The half-automated version drops it twice and picks it back up.

What to do instead

The move isn't "automate more". It's to look at where the seams fall.

A workflow with one clean handoff at the end usually beats a workflow with three scattered through the middle — even when the second one automates more steps overall.

So: automate a whole contiguous block rather than the easy steps wherever they happen to sit. Easy steps are easy because they're small. And small steps are exactly the ones whose removal saves the least and costs the most.