One doc, four readers

Every big document I've ever shipped failed somebody. The executive got forty pages when they needed four sentences. The engineer got four sentences where they needed the edge cases. Same doc, same day, opposite failures.

One document aimed at everybody lands on nobody.

Old writing advice says to picture one reader (the avatar) and write to them. Good advice, hiding a problem: an RFC never had one reader. It always had four, and they were never going to be served by the same page.

The fix I've landed on is almost embarrassingly simple: write it four times. Google Docs grew tabs, and tabs turn out to be the feature the RFC was waiting for. One document, one title, four doors:

• The idea. One page, plain language. What we're doing and why anyone should care. For the person who needs exactly that and nothing more.
• The executive tab. In the know, not in the weeds. The decision being asked for, what it costs, what it risks, when it lands. It reads in three minutes, because three minutes is the budget.
• The implementation tab. For the person building it: the decisions we made and why we made them. Not a spec dump, a decision record.
• The kitchen sink. Everything. References, transcripts, dead ends, the why behind the why. Nobody has to read it. Somebody will be very glad it exists.

Reading is the expensive part. I've priced this elsewhere: a person's tokens cost about a thousand times the model's, and an executive's attention runs at the top of that curve. Making them excavate your implementation details to find their decision is spending their most expensive tokens on your least relevant content. (I built a calculator for exactly this feeling. It's called Human Tokens.)

The old excuse was effort. Writing four versions of anything was a tax nobody paid, so we shipped the forty pages and called the reader lazy. That excuse died. Generation is nearly free next to the reader's time. I draft the executive tab from the implementation tab in about a minute, with the AI partner I build everything with. Which flips the ethic: when generating is cheap, curating is the work. Anyone can hand you fifty pages. The gift is the five your reader actually needed.

The worry has a fashionable name, and the name blames the wrong party. Most of what gets filed under AI slop is human laziness: someone let the model generate and skipped the part that was always theirs to do. The model will hand you fifty pages. Shipping all fifty is a person's choice, and a lazy one. Four tabs is the opposite. It's one idea cut four ways by someone who did the cutting, who decided which five paragraphs each audience needs and dropped the rest. That deciding is the whole job now, and it's mine, not the model's. Nobody did it before because four versions meant four times an engineer's most expensive hours, and no one was paying that. The idea was always right. AI only made it cheap.

A reader's focus is the rarest input you get. When someone gives your document twenty minutes of it, that's a gift, and the tabs are how you honor it.

So: one doc, four readers. It was always four documents wearing one title.