Keep AI Away From Writing That Helps You Think
An OpenAI product builder automates status updates freely, but writes strategy drafts herself, using AI only in the middle for research and pushback.
Lenny's Podcast: Product | Growth | Career · AI’s third era: the rise of persistent AI coworkers | Tara Seshan (OpenAI’s product lead)Tara Seshan, who works at OpenAI and whose strategy briefs were called iconic in the interview, keeps a strict rule for working with AI. She divides everything she writes into two piles, and only one of them is allowed near the machine.
The first pile is writing as thinking. That is a brief about why to build a product, why to take a certain strategy, or a bold or contrarian view on where things should go. The second pile is writing as reporting. That is a status update on what the team did that week, a launch plan, or a summary translated from one format to another.
Her method is simple. Automate the second pile as much as possible, and never automate the first. She said both extremes are wrong: never using models for writing, and always using them for writing. The distinction is what matters.
For thinking work, she does the hard sequence herself. She outlines the argument, turns it into prose, cuts it, edits it, and iterates until her ideas line up. She does not have the model generate the first version, and she does not rely on it to polish the prose, because that drafting and revising is how she gets her thinking straight.
She allows AI only in the middle. Once she has started the draft herself, she might have it research a point, pull in data, or push back on her ideas. As she put it, "I start myself and I end myself."
Her personal rule of thumb is time. She said that if she is going to ask people to read a document, she has to put in at least as much time writing and rereading it as she expects them to spend reading it. She applies the same math to meetings, prepping for the total amount of attendee time a meeting will consume.
For anyone trying it this week, sort tomorrow's writing before you open a chatbot. If the document is how you will figure out what you think, open a blank page and write the ugly first draft yourself.
Questions people ask
How can I tell whether something I have to write is thinking or reporting?
Ask what the document has to decide. If it needs to persuade someone to build, fund, or choose something, or to clarify what you believe and why, treat it as thinking and draft it yourself. If the decision is already made and you are just informing, tracking, or reformatting, treat it as reporting and automate it.
What if I want AI to draft a first version because blank pages are painful?
That rough start is the point. Seshan argues that outlining, writing, cutting, and editing is what lines up her ideas, so skipping it skips the thinking. Use AI in the middle for research, data, or pushback, but keep the first and last passes human.
Does she ever use AI to make her writing sound better?
Seshan says she does not, because she does not find that it improves her prose. She rereads and revises her thinking pieces herself.