Inbox Zero Is an Outbox Problem: Triage Sorts What Arrives, Not What You Send

The Treadmill

Inbox Zero is the productivity dream almost nobody keeps. You read the posts, you spend a Saturday archiving, and by Monday afternoon there are forty-seven new messages and you are exactly where you started.

The tooling response has been triage: AI that reads incoming mail and sorts it by urgency, so you spend your attention in the right order. Urgent, important, FYI, newsletter. Superhuman calls it Split Inbox, Shortwave calls it Bundles, SaneBox has built a whole business on it.

It works. Ranking inbound mail is a solved-enough problem, and the tools that do it save real time.

It is also solving the half of email that was never going to end your quarter.

Which Half of Email Actually Hurts

Run the tape on the email disasters you have personally witnessed. Not the annoyances — the disasters.

Every one of them was outbound.

The engineer who pasted a live key into a message to a contractor. The account manager whose "as discussed, here's the revised pricing" went to the lookalike thread with a different customer. The VP whose reply-all included the internal negotiating position. The candidate whose follow-up read so much like a chatbot that the hiring manager stopped reading at the second paragraph.

No triage system sees any of those, because in each case the message had not been received by anyone yet. It was sitting in a compose window, one gesture away from being permanent, and every tool in the stack was pointed the other direction.

Inbound failure costs you time. Outbound failure costs you the thing the email was about.

Why Presend Does Not Ship Triage

Presend is a gate, not an assistant. It does not summarize threads, draft replies, or categorize your inbox — and those are not gaps waiting to be filled in a later release. They are the deliberate shape of the product.

The reason is focus of the boring kind. A tool that sits in the compose window has exactly one moment of leverage: the instant between deciding to send and sending. That moment is short, it is interruptible exactly once before it becomes annoying, and it is the only point in the whole email lifecycle where a mistake is still free to fix.

Spending that moment well takes the entire product. Every feature that reads your inbox instead of your draft is attention pointed away from it.

What Happens at the Send Gesture

The send button never stops working. Presend intercepts the gesture — a click, or the platform's own send keystroke — in the capture phase, evaluates the draft, and shows a confirmation sheet only when it found something classed as risk. Warnings and observations do not block anything; they sit in the panel beside your draft while you write.

What it looks for:

Risk, which stops the send: credentials and API keys, including ones the sentence labels rather than ones we pattern-match; numbers shaped like Social Security or payment-card values; phrases your company has banned outright.

Warnings, which do not: "attached" with no attachment; "see the link" with no URL; an empty subject; a company named in the subject but nowhere in the body; phrasing that reads as model-generated, with a suggested rewrite.

Observations: how many external recipients are on the message, listed by address, before rather than after.

Each finding underlines the exact span that caused it. That matters more than it sounds: a dial cannot be acted on, a highlighted sentence can.

The Two Questions Scoring Answers

Alongside the rules, two independent engines ask two different questions about the draft.

Is this a model? A deterministic catalog of AI tells, requiring no history and no setup. It works on your first day.

Is this you? A voice comparison against your own sent messages, using Burrows's Delta over character trigrams, function words, and style scalars, each z-scored against your own variation rather than an abstract norm.

A message can sound like you and still be slop. It can be slop-free and still not sound like you. Kept separate, neither engine has to be tuned aggressively, and each covers the other's near-misses.

The voice profile builds itself from what you send, because standing at the gate means every sent message is a labelled sample of your real voice on that platform. It runs locally in the content script, the corpus never leaves the machine, quoted reply chains are stripped before storage, and it abstains below five samples or four hundred words rather than guessing.

Your Own Rules Are the Point

Built-in checks cover what is true for everybody. The checks that catch your mistakes are specific to your job, so you write them in plain English and give each one a severity:

• "Block any message containing a customer's contract value."
• "Warn me when I commit to a date without naming the day."
• "Flag mentions of unreleased product names to anyone outside the company."

On Team, those rules are shared org-wide, so the policy lives in one place instead of in a document nobody rereads. Rules can also call a webhook or your own API when the check needs data Presend does not have — a CRM lookup, a deal-stage check, an internal allowlist.

Where It Runs

Gmail, Outlook Web, Slack (including plain-Enter sends), LinkedIn DMs and posts, X, and every other textarea and contenteditable on the web, found structurally rather than by a per-site selector. Notion, Jira, Zendesk, HubSpot, a SaaS comment box nobody wrote an adapter for.

Rules are filtered by channel, so email checks stay in email, secrets and slop run everywhere, slop escalates to a hard stop on public posts, and documents are advisory with no gate at all. Below a per-channel word floor — twenty-five for chat, twelve for email — Presend stays silent, which is the difference between a tool people keep and a tool people mute.

What It Costs

Free forever covers twenty-five AI checks a day, with local rule checks and voice scoring unlimited because they run on your machine. Pro is three dollars a month: unlimited checks, custom rules, voice-preserving rewrites, send-guard, personal analytics. Team is twenty-nine dollars per user per month for shared rules, the Custom Actions pipeline, webhooks and the full dashboard.

Presend does not store message bodies — only a one-way hash that cannot be reversed into your text. The voice corpus stays on your device.

Keep Your Triage

If AI triage saves you thirty minutes a morning, keep it. This is not a competitive claim; Presend runs alongside Superhuman, Shortwave, SaneBox and Gmail's own tabs without touching any of them.

Just notice what none of them are watching. They are all reading the mail that arrived. The message that can still be stopped is the one in front of you.

Add Presend to Chrome — Free →