Journal · The AI Architect · 2026-08-29
Break a big goal into small, clear parts before you hand it off. You can't oversee what you haven't divided up.
Break a big goal into small, clear parts before you hand it off. You can’t oversee what you haven’t divided up.
The tension isn’t really about breaking work into steps. Most people know they should do that. The tension is that dividing a goal forces you to admit what you don’t yet understand about it — and it’s tempting to hand off the whole vague shape of a problem rather than sit with the discomfort of figuring out its parts first.
Name what you actually know
When a goal feels big, it usually means it’s still fused together — the what, the how, and the why are all tangled into one impression. “Improve customer onboarding” or “get the new hire up to speed” sounds like a single thing, but it isn’t. It’s a bundle of smaller decisions, each with its own shape and its own way of being checked.
The work of dividing a goal starts with an honest inventory: what do I actually know about how this gets done, and where does my knowledge run out? That second part matters more than people expect. The moment you hit the edge of your own understanding is exactly the moment you’re tempted to just say “handle onboarding” and walk away. But that’s also the moment worth pausing at, because whatever you don’t understand, you also can’t inspect later. You’ll either have to trust blindly or hover anxiously — and neither is oversight, they’re just two flavors of not knowing.
Try this: before delegating anything, write down the goal in one sentence, then list every decision someone would need to make to get there. Not the tasks — the decisions. “Improve onboarding” might break into: decide what “success” looks like for a new customer, decide which step causes the most drop-off, decide what to change first, decide how to measure whether the change worked. If you can’t list the decisions, you’re not ready to hand the goal off yet. You’re ready to go find out more.
Divide by what can be checked, not by what feels tidy
A common mistake is dividing a goal into parts that are easy to describe but impossible to verify. “Make the report better” split into “improve the intro,” “improve the middle,” “improve the conclusion” looks like division, but none of those pieces tells you what “improve” means or how you’d know it happened. You’ve split the goal without making it checkable, which means you still can’t oversee it — you’ve just relocated the vagueness into three smaller vague boxes.
Good division creates parts with visible edges. Each piece should have a clear starting condition, a clear stopping condition, and some way — even a rough one — to tell if it went well. “Rewrite the intro so a first-time reader understands the problem in two sentences” is checkable. “Improve the intro” is not.
Try this: for each part for you’re about to assign, ask “how will I know this is done, and how will I know if it’s done well?” If you can’t answer both questions in a sentence, the part isn’t finished being divided. Cut it smaller, or define it more sharply, until the answer comes easily.
Let the parts tell you who should own them
Once a goal is genuinely divided, something useful happens: the parts start suggesting their own owners. A piece that requires judgment calls about customer psychology goes to someone who’s spent time with customers. A piece that’s mostly mechanical — updating a template, checking a list — can go to someone newer, because the checkable edges you built in mean they can’t quietly drift off course without you noticing.
This is the real payoff of dividing before delegating. It’s not just that the work gets organized. It’s that oversight becomes possible without becoming controlling. You’re not looking over someone’s shoulder the whole time; you’re checking in at the edges you already defined. The division does the watching for you.
If you skip this step, oversight tends to curdle into one of two things: either you disengage completely and hope for the best, or you hover over every step because nothing was ever specific enough to check at a distance. Neither is good management. Both come from the same root cause — a goal that was handed off whole.
Dividing a goal well takes longer up front than just assigning it. That’s the trade people resist. But the time spent there is what makes the rest of the delegation actually work, instead of just looking like it did.
There’s more in the book about how to divide goals when the work itself is unfamiliar to you, and how to adjust the size of the pieces as someone grows more capable. But the starting move is the one above: know what you know, make the parts checkable, and let the parts tell you who should hold them.
Go deeper. The full method is in The AI Architect. New here? Start with the free companion pack, or explore the series.