Email is Dead. Long Live Email.

The first email client I used in earnest was Outlook 2003. It had a preview pane, a folder tree, a list view, and a status bar. The Gmail tab open in my browser right now has a preview pane, a folder tree (renamed "labels"), a list view, and a status bar. The only thing that has changed in twenty-three years is the keyboard shortcut for archive.

In every other category of software, an interface that has not been redesigned in twenty-three years would be considered a failure of imagination. In email it is considered the natural state of things. We have decided, collectively, that the inbox is finished — that this rectangle of unread bold and read regular is the optimal shape for the central nervous system of work.

It is not the optimal shape. It's just the one that worked.

That said, the email client as a product category is on a 3 to 5-year clock. Email the protocol survives. It is the only universal addressable surface that every colleague, client, counterparty, vendor, board member, and government agency on the planet agrees on. Email the interface, the thing you open and stare at and triage from, is not going to survive contact with what comes next. The replacement is already here in early form.

Backing up, let me start with the numbers, because the numbers are the part nobody is arguing with.

361 billion emails are sent per day. The average knowledge worker checks email 74 times a day. The average attention span on a screen is roughly 47 seconds. 28 hours per week (across all measured studies, the number lands between 25 and 32) are spent on email by the average information worker. 82% of phishing emails sent in 2025 were machine-generated. The volume is exploding, the attention is collapsing, and the interface has not been redesigned in two decades. This is a system under structural strain. Something gives.

For most of the last decade, the answer the category gave was: make the interface faster. Superhuman built a beautiful product around shaving milliseconds off triage. Shortwave and Spark built thoughtful redesigns of the list view. HEY rejected the entire premise of the inbox and rebuilt the metaphor. These were honest, well-built products. They optimized the right side of the equation if you believed the right side was the manual interface.

I do not believe that anymore. I think the manual interface is the problem.

Nobody opens Gmail because they want to read email. They open Gmail because they want to do the thing the email is about. Write the reply, follow up with the vendor, get the answer they need to move the deal forward, send the contract back, schedule the call. The reading is a tax on the doing. The triage is a tax on the deciding. The folder structure is a tax on the remembering. The interface is a tax on the work.

A faster interface lowers the tax. A different shape of product eliminates it. AI solves this.

The shape of product that eliminates it is an agent with continuous existence. Not a chatbot you open. Not a sidebar in your client. Not a Chrome extension that drafts inline. An agent that lives between conversations, has its own identity, has its own inbox, watches the threads you tell it to watch, files what arrives, runs the routines you set, and surfaces only the things that actually need you. The substrate is email because email is the only universal protocol, but the experience is not "opening email." The experience is the work being finished before you would have started it.

I want to give the strong version of the counter-argument, so as not to be too hand wavy here. The strongest version is: email is too important and too personal to delegate. Every meaningful relationship in your professional life passes through your inbox. You can't outsource that. People will notice. The voice will go off. The judgment will fail at exactly the moment that matters.

I have heard this argument many times in the last year. I think it is right about the magnitude of the risk and wrong about the conclusion. The conclusion that follows from "email is too important to delegate" is not "do not build agents in the inbox." The conclusion is "build agents that earn the delegation a single rep at a time, with constraints that make the worst case visible and the best case repeatable." The hard constraints we ship with at Deck (e.g., the agent never reads your inbox, only what you forward; the agent never sends to anyone but you; the agent never installs anything) exist for exactly this reason. The delegation is asymptotic. It compounds. The user who forwards one email today forwards five tomorrow because the first one came back done. The user who forwards none never gets a second chance.

The other strong counter-argument is: AI is not yet good enough to do the actual work, only the triage. This was true for a long time and is no longer true. Multi-step reliability crossed a threshold in the last twelve months. The models can now hold a plan across a four-message exchange, catch a discrepancy three threads back, draft a reply that does not need to be rewritten, and decide which of the day's incoming items genuinely require human judgment. The category I am building in is the category that exists because that threshold moved.

What does this mean for your week, if you are an operator reading this?

It means that the activity you have spent the last decade trying to optimize - your own inbox triage - is on its way out. The keyboard shortcut you mastered to archive faster is the last optimization on a workflow that is about to be obsoleted. Stop measuring inbox zero. Start measuring how much work happens to your inbox while you are not looking at it. The next generation of operator productivity metrics is not the speed at which you process email. It is the percentage of email that processes itself.

It means that the architecture of your day shifts. The 6am inbox scroll dies. The post-meeting "let me clear my inbox" twenty-minute ritual dies. The end-of-day "I have to clear out before I close my laptop" hour dies. What remains is the work that genuinely requires your judgment: the negotiations, the relationships, the decisions, the calls. The portion of your day that was tax becomes the portion of your day that is work.

It means that the email client as a product category enters terminal decline, even though email as a protocol persists forever. Gmail will still exist. Outlook will still exist. They will become what the file system became after the desktop metaphor stopped being how people thought about their work — an underlying substrate that experts touch and most people don't think about. The actual interface for most knowledge workers in five years will be the agent. The agent will reach into the inbox when it needs to. The user will mostly not.

I will be wrong about the timeline. The wildcard / slow part is likely to be the trust infrastructure - the permissions, the audit, the user's lived experience of an agent that does not embarrass them.

I will not be wrong about the direction. The inbox is not being optimized. It is being obsoleted. We are inside the change. The companies that win the next decade are the ones that figured out what to ship to the inboxes of people who have stopped opening them.

Anyhow, that is the company I am building. The product is called Deck. The shape is an AI sidekick with its own email address. The thesis is the one in this essay. The first hundred customers told us we were on to something. The next ten thousand will tell us how wrong we still are about the details. We will ship through it.