Email isn't broken — composition is

Two decades of email products reorganized the inbox. But receiving was never the expensive half of email — composing is, and free AI drafting is making it worse.

5 min read

Every few years a product arrives with the same diagnosis — email is broken — and the same class of cure: a better inbox. Filters. Labels. Snoozes. Digests. Priority sorting. Two decades of genuine ingenuity, aimed almost entirely at one half of the medium: the half where mail arrives.

The diagnosis was wrong the whole time. The expensive part of email was never receiving. It was composing. And the first tools to actually touch composing — AI assistants — are making it worse.

The fixes all landed on the same half

Credit where due: inbox overload is real, and triage tools genuinely help. A good filter saves an hour a week. A digest turns forty notifications into one. Snoozing keeps Thursday's problem out of Tuesday's attention. None of that is disputed here — if your inbox is a landfill, triage is the right tool for the landfill.

But notice what every one of those tools has in common: it reorganizes what arrives. Triage treats email as weather — volume that happens to you, to be sorted, deferred, and survived. It never touches the other half of the medium, the half where someone decides what to write.

Email is not weather. Every message in the pile was composed by a sender who either did the work of being answerable, or shipped that work to you.

Composing was always the expensive part

Count the costs honestly. Receiving a message takes seconds of triage and minutes of reading. Composing a good one costs everything behind it: the hour of context it stands on, plus the editorial work of compression — deciding what the ask is, what this particular reader needs, and what to leave out. A good email is a transfer of labor from reader to writer. The sender compresses once so the reader does not have to excavate.

Which is why most of what makes an inbox miserable is not volume. It is composition failure arriving at scale: the forwarded thread with "thoughts?", the five confident paragraphs with no ask, the question you have to read twice to find. A triage tool can sort those messages, snooze them, even summarize them. It cannot supply the missing ask, because the sender never decided what it was.

Free drafting makes it worse

For all of email's history there was one crude control on this: writing is work. Effort was a bad editor, but a real one. The blank page rate-limited volume and forced at least some compression before sending.

AI assistants removed that control. Drafting is free now — fluent, grammatical, confident, endless. And because composing was never the typing but the deciding, free drafting does not fix bad email; it scales it. A model can expand a vague thought into five plausible paragraphs. It cannot decide what you are asking for, what you actually believe, or what you are willing to commit to. That deciding is the whole job, and it is exactly the step "just draft something" skips.

Follow this to its endgame and you get an arms race of fluency: your assistant inflates a thought into a page, my assistant compresses the page back into a line, and two models negotiate prose while both humans skim. There is more text than ever and less intent than ever, and nobody has committed to anything.

What composing for the recipient means

Three moves, none of which a model can make on your behalf:

State an explicit ask. What do you want — a decision, an opinion, a review of something specific — and how urgent is it, honestly. A reader who cannot tell what shape of answer you need will default to the cheapest one: none. There is a practical guide to this in How to ask for a decision.

Write for a cold reader. The recipient was not at your desk for the hour of context. Give them the background the answer depends on, not the journey you took to get there. What you tried and discarded is desk work, and it stays at the desk.

Label your assumptions. Some of what you send, you stated. Some of it, your assistant inferred. A reader who can tell the difference can challenge the right layer — "that inference is wrong" is a cheap correction; untangling who actually said what, three replies later, is not.

Most of this is subtraction. Composing for the recipient mostly means deleting what serves you and keeping what serves them.

Where briefings fit

A briefing is this discipline made structural. Instead of hoping the sender remembers the moves, the format demands them: an ask with a response shape and an honest urgency, a TL;DR, a context brief written for a cold reader, assumptions labeled stated by sender or inferred by sender's AI, options considered, decisions made, open questions, and a note from the sender labeled with exactly who wrote it. The full anatomy is in What is an AI briefing?.

RelayLink exists to carry that unit. You think the problem through privately with your own assistant; the session never leaves your side. When a real person needs to weigh in, your assistant composes the briefing, you review the exact package before anything is released, and the recipient gets something a busy person can answer — delivered as ordinary email, no account or assistant required on their end.

Two honest notes. A briefing costs more to send than "thoughts?" — deliberately; the point is to move the cost back to the sender's side, where email always worked best. And RelayLink does not fix your inbox. It never touches your inbox at all — it is a separate channel — so your triage tools keep their job. It changes what you put in other people's.

Drafting is free now. Deciding is still yours. If you want the structure that keeps those two separate, connect your assistant — the person you send to needs nothing but an email address.

Frequently asked questions

Why do people say email is broken?
The complaint is usually about receiving — too much volume, too little signal — and that part is real, which is why filters, snoozes, and digests genuinely help. But those tools only reorganize what arrives. The costlier failure sits on the sending side, in messages composed without a clear ask that transfer the work of understanding to the reader.
Doesn't AI make email better by writing it for you?
AI makes drafting fast, which is not the same as making composition good. Composition is deciding what you are asking, what a cold reader needs to know, and which claims are yours versus your assistant's inferences. A model can propose those decisions, but only the sender can own them — skipping that step produces fluent volume without intent.
What does composing for the recipient actually mean?
It means stating an explicit ask with the shape of answer you need and an honest urgency, giving a cold reader the context the answer depends on rather than the journey you took, and labeling which assumptions you stated versus which your AI inferred, so the reader can challenge the right thing.