I Love My InboxAnd I Want to Share It

The inbox started as a text-first personal space. A direct simulation of physical mail. You chose an address as a nickname and used it to exchange private messages on the internet.

This is illustrated very well in the movie You've Got Mail¹. At the beginning, we see the two characters sneak toward the computer, turn it on, and start messaging each other through their personal inboxes, almost like a secret activity.

¹ You've Got Mail, directed by Nora Ephron, Warner Bros., 1998. Released in the United States on December 18, 1998.

That was 1998. Open that scene next to your inbox right now and the interface is almost the same: a list, a sender, and a subject. But how we use it has changed substantially.

The inbox became a to-do list, a document archive, a recipe box, an authentication layer, a business memory, and so many other things. But most of all, it became a place where we take actions instead of only reading messages.

The Everything Interface

It's fascinating how the same interface keeps surviving and thriving. It became our mental model for dealing with multiple things. It bends to almost any department and task. Look at Linear and you can see the strong influence of the inbox in their product as an orchestration interface.

But for us to unleash the potential of this interface, we need to leave the personal message metaphor behind. The inbox is still designed around an old assumption: one inbox belongs to one person.

You should be able to create as many inboxes as you want, guaranteeing the right permissions for each one, and then share each one with whoever needs it: another person, a piece of code, or an agent.

Take shopping. I could create one inbox and use that address every time I buy something online. Receipts, shipping updates, promotions, and coupons would all arrive there instead of polluting my personal inbox.

Because the inbox is programmable, I could write a function that pulls out every coupon, and another one that tracks what I spent this month. I could tell it to only email my personal address when something actually needs a decision.

And how incredible would it be to email that same inbox and ask an agent to find a chair for my apartment? Because of the history of my past purchases, the search is probably way more accurate to my style, budget, and brands. The receipt for the chair would land in the same place as the conversation that chose it.

For your house, an agent could help you contact your landlord, organize a repair, compare internet providers, or keep track of recurring bills.

As a team, you could create a shared space to keep more people in the loop and teach agents how to draft, organize, and categorize your emails, condensing business knowledge just by using it.

And all these inboxes could communicate with each other because they are just one email away.

The open inbox

We already see a version of this with WhatsApp in Brazil. For many businesses there, WhatsApp is not just messaging. It's customer support, sales, scheduling, payments, and operations. In a survey of 323 Brazilian retailers, 96% used it to talk to customers and 82% treated it as their main communication tool.²

But WhatsApp is a platform. Email is a protocol.

Mike Masnick called this "protocols, not platforms": when the thing in the middle is a protocol, nobody has to ask permission to build on it, and nobody can change the rules on you later.³ WhatsApp gets to decide what a business is allowed to do with it. SMTP hasn't been able to decide anything since 1982.⁴

My house inbox can email the landlord, and the landlord doesn't have to install anything. He replies from Outlook. The thread still works.

The primitive foundation

So why is everyone building a brand new interface for agents instead? Look at the primitives of the email clients:

1. Asynchronous. Agents are asynchronous, and the inbox already works very well for asynchronous collaboration. You send a request, continue with your day, and return when there is a reply or when a decision is needed. The agent does not need to simulate a live conversation all the time.
2. Auditable. Agents also need to be audited. We need to know what they received, what they did, what they changed, and why they made a decision. The inbox already keeps this history. The messages, attachments, replies, and approvals create a record of the work without requiring a completely new system.
3. Authenticated. It knows who you are, who sent the message, who has access to the conversation, and who can take action. We do not need to recreate all of this every time we build an interface for an agent.
4. Interoperable. It already connects people and companies running completely different software. Agents need exactly the same thing, and they don't need a new protocol to get it.

We can shape it, change it, and adapt it around the conversation. And it can evolve with the history of how you write, how you answer different types of emails, and how you label things. All these actions can easily feed your agent's context and help it prepare things for you.

The inbox is the perfect orchestration software because it is not starting from zero. It is already a place for identity, documents, memory, actions, collaboration, authentication, and automation.