Guide
Should you write the first draft?
Three questions to ask yourself before you let AI start.
“I write entirely to find out what I’m thinking.”
Joan Didion
Not to express thinking. To discover it.
AI can write your first draft in seconds. But the draft was never about the document. It was about what the struggle forced you to see. The first draft is where half-formed instincts become positions. Where you discover what you actually believe, not just what sounds right.
Cognitive psychologists call this desirable difficulty: the kind of friction that feels inefficient in the moment but builds lasting capability. AI removes that friction. Which is sometimes exactly what you want, and sometimes exactly what costs you.
The question is not whether AI can draft. It can. The question is whether you should.
The Framework
Three questions before you delegate
Each question targets a different reason the draft matters. They are not a checklist. One clear answer is enough to decide.
Do I have the competence to evaluate what AI produces?
Draft yourself when
- This is new territory and you are still building the skill
- You would not recognize errors in the output
- You could not explain the result without leaning on AI
Example: You are writing your first financial model narrative. You do not yet know what “good” looks like well enough to judge whether AI got it right.
Let AI draft when
- You have done this enough to catch mistakes on sight
- The format is routine; the judgment is yours
- You could rewrite it from scratch if the output were wrong
Example: The weekly status update you have written for two years. You know exactly what it should say. AI saves you the typing.
Do I need to discover what I think?
Draft yourself when
- Your position is not clear yet
- You are working through complexity, not documenting a conclusion
- The writing is the thinking
Example: Your quarterly strategy review. You have the data, but you have not decided what it means. The draft is how you find out.
Let AI draft when
- You know exactly what you want to say
- The structure is settled; you are documenting, not discovering
- Speed is the constraint, not clarity
Example: A project brief where the decisions are already made. The thinking happened in meetings. Now you need it written down.
Does my voice matter here?
Draft yourself when
- The reader expects you, not a document
- Your judgment and perspective are the value
- Being distinctive matters more than being efficient
Example: A recommendation letter, a board memo with your name on it, a piece of thought leadership. The reader is buying your thinking, not a polished format.
Let AI draft when
- The format is standard and the reader expects consistency
- Personality would be a distraction
- The content matters more than who wrote it
Example: Internal documentation, compliance reports, templated client communications. The value is in the information, not the author.
The Middle Ground
Most real decisions are not binary
The framework above clarifies the extremes. In practice, many tasks live between them. A few hybrid approaches that preserve the thinking while using AI where it helps:
Outline first, then delegate
Write the structure and key arguments yourself. Let AI expand sections where you have already done the thinking. You keep the architecture; AI handles the construction.
Draft rough, then refine with AI
Get your thinking down in raw form. Use AI to restructure, tighten, or reformat. The discovery happened. Now let the machine handle polish.
Draft in parallel, then compare
Write your version. Have AI write its version. Compare the two. The gap between them reveals what you know that the model does not, and what the model caught that you missed.
The Trap
“I’ll learn from editing AI’s draft.”
There is some truth here. Editing a draft does surface thinking. You react, disagree, reshape. That is real cognitive work. But it is a different kind of work than drafting.
Editing is responsive: you evaluate choices someone else made. Drafting is generative: you confront the blank page and make choices that reveal what you actually think. Editing builds judgment about quality. Drafting builds the capacity to produce it.
The danger is not that editing is useless. It is that editing feels enough like thinking that you stop noticing what you skipped.
AI perfects the output. It eliminates the process that made the output yours. The draft you never wrote is the thinking you never did.