Most people use AI to execute. Write this code. Draft this email. Summarise this document. The prompt is the plan, and the output is the result.
That works for small things. But for anything with scope — a feature, a project, a product — execution without planning is just motion. You end up with a pile of outputs and no structure connecting them.
Start Before the Prompt
The best use of AI isn’t the final step. It’s the first one. You have a rough idea — maybe a voice note captured on a walk, maybe a sentence scribbled in your inbox. Something that isn’t a plan yet but could become one.
Hand that to an AI and ask it to decompose. Not to execute — to break down. What are the pieces? What depends on what? What’s the first concrete action?
The result isn’t a finished plan. It’s a draft you can react to. Rearrange, cut, reprioritise. That’s faster than building the plan from scratch, and it surfaces things you’d have missed until you were halfway through.
The Workflow
In Task Register TR-1, this is a concrete loop:
Capture. A voice note in Signal Recorder SR-7, a quick text capture, a thought dropped into the inbox. No formatting, no structure. Just the raw idea.
Process. AI reads the inbox item. It asks: is this actionable? Is it a project or a single task? What’s the context? It proposes a project with tasks, estimates energy, suggests tags. All of this happens as a proposal — nothing moves until you approve it.
Review. You look at what the AI proposed. You reorder. You defer the things that aren’t ready. You promote the things that are. This is pure judgment work — the kind AI can’t do for you and shouldn’t try to.
Execute. Now the agent picks up tasks from the backlog. The plan is already in place. The decisions are already made. Execution is the easy part.
Decomposition vs. Prioritisation
AI is remarkably good at breaking things down. Give it a vague goal and it will produce a reasonable task list — often more thorough than what you’d write yourself, because it doesn’t skip the boring parts.
But it has no sense of what matters. It treats every subtask as equally important because it has no stakes in the outcome. That’s where you come in. The human role in planning isn’t to list every step — it’s to decide which steps matter now and which can wait.
A weekly review makes this concrete. You look at every active project. You ask: is this still relevant? What’s stuck? What’s the single next action? The AI can prepare the review — pull the data, flag stale projects, surface tasks with no recent activity. But the decisions are yours.
Plans Change. That’s the Point.
A plan created with AI isn’t sacred. It’s a starting position. The value is in having one at all — something to react to, adjust, and override as you learn more.
The alternative is prompting your way through a project one step at a time, discovering the shape of the work as you go. Sometimes that’s fine. But for anything that matters, thinking before doing is still the move. AI just makes the thinking faster.