Which tool to choose
You don't need a special tool: it's a method that works with any assistant. In fact, it's worth having the AI itself help you decompose the task before executing it. If the sub-tasks are many and independent, you can tackle them in separate chats to keep them light; if they're few and connected, a single conversation managed in stages is fine.
How it's done
- Focus the final goal and get help breaking it down. The operational syntax:
I have to [big goal, described in one line]. Help me break it into an ordered list of clear and as-independent-as-possible sub-tasks, from first to last. For each one, indicate what's needed as input and what it produces as output. Don't execute anything yet: for now I just want the list.
- Review and adjust the breakdown: this is the moment when you decide the structure of the work.
- Execute one sub-task at a time, giving the AI only the one it has to work on.
- Pass the result of one step as the input of the next, until the work is complete.
A concrete example
Paolo had to organize the opening of his shop and felt paralyzed by the scale of it. He asked the AI to break down the task instead of "organize the event for me". A list came out: define budget and date, make the guest list, choose catering, prepare the social communication, plan the setup. He approved the structure, then worked one piece at a time: first budget and date, and once those were set, the rest fell into place. What seemed a boulder became five manageable tasks, each one closed before moving to the next.
When it does NOT work (and how to fix it)
If the sub-tasks depend too much on each other
If every piece needs all the others to be done, the breakdown doesn't help. The fix is to redesign them more independent, or to explicitly pass the output of one step as the input of the next, so the dependency becomes a clear handoff instead of a tangle.
If you get lost among the various pieces
With many sub-tasks it's easy to forget where you are. Keep a small document that tracks the state of each one — to do, in progress, done — so you always have the map of the work under control.
If the AI redoes everything instead of the single piece
Sometimes it tends to tackle the whole task again. Tell it clearly: "work only on sub-task number three, ignore the others for now". Circumscribing the request keeps it focused on the right piece.
A tip from someone who really uses it
A breakdown done well is already half the work. Invest a few minutes in having it give you the list of sub-tasks and reviewing it calmly before starting: a sequence of clear steps is enormously easier to carry through than a single monster tackled in a rush. Often, seeing the pieces lined up, you also realize what you can skip or simplify.
Frequently asked questions
Is it the same thing as prompt chaining?
It's the same principle applied to big tasks: breaking up and chaining the steps, using the output of one as the input of the next. Here the emphasis is on the initial decomposition of a broad goal into a structure of sub-tasks, even before executing them.
Into how many sub-tasks should I divide?
Into as many as are needed for each to be a single, clear and manageable thing. There's no magic number: what matters is that no piece remains a boulder in its turn. If a sub-task is still too big, break that one down too.
Is it worth using separate chats for each sub-task?
For independent pieces, often yes: it keeps each conversation light and focused. For tightly connected pieces, a single chat managed in stages is more convenient. It depends on how much the sub-tasks need to "see" each other.
Doesn't giving everything at once save time?
No, and it's the illusion that costs the most. A huge request produces a confused and approximate result on everything, which you then have to dismantle and correct piece by piece: the time "saved" in asking you pay back with interest in fixing. Decomposing costs a few minutes more at the start and gets you to a better, already-checked result. The haste to ask everything at once is the longer road.