The New Moat Is the Session Token

AI can rebuild almost any app. What it can't rebuild is the connection between a user and their bank, phone number, or work apps, and the companies holding those keys are the new moat.

If you're building a startup in the age of AI agents, here's an uncomfortable truth: almost everything you build can be copied. Quickly.

The app, the interface, the features, even the clever workflow you spent six months on. An AI coding agent can rebuild most of it in a weekend. That's not an exaggeration anymore. It's just how software works now.

So what's left that can't be copied?

My answer: the session token. More specifically, a session token established close to the metal, one that ties a user to a service and can't easily be moved somewhere else.

First, what's a session token?

When you log into something, the service gives your app a kind of digital key. It's a long string of characters that says "this person already proved who they are, let them in." That key is a session token. It's why you don't have to type your password every time you open an app.

Here's the important part. That key usually belongs to whoever set up the connection. If a middleman set it up, the middleman holds the key. You don't. The app you're using doesn't either.

What "close to the metal" means

Engineers use "close to the metal" to describe things that sit near the bottom of the technology stack, near the actual hardware, networks, and core systems. The further up you go, the closer you get to what the user sees: the app, the buttons, the personality.

Picture a stack of layers:

  • At the bottom: the bank, the phone network, the actual service.
  • In the middle: the plumbing that connects apps to those services.
  • At the top: the app or AI assistant you talk to.

The top layer is where most startups live, and it's the layer that's getting commoditized fastest. The bottom layer is owned by giants. The interesting money is moving to the middle, to the companies that hold the connection between the user and the service.

Three companies make this concrete.

Example one: Linq and your phone number

Linq builds the plumbing that lets developers send and receive iMessages and text messages. If you've ever texted an AI assistant like it's a friend, with blue bubbles, typing indicators, and reactions, there's a good chance something like Linq was underneath it.

Investors have noticed. Linq raised a $20 million Series A in February, and a16z just announced it is leading Linq's Series B. Many of the fastest-growing agent companies use Linq to reach their users.

Here's why that's powerful. When you start texting an assistant, the phone number you text runs through Linq. Your whole relationship with that assistant lives in that thread: the history, the context, the habits. The assistant's memory and everything built on top of it is tied to that connection.

Now try to leave. You can't take that thread with you. A new assistant means a new number, a new conversation, and starting over. The connection Linq set up is the part that sticks.

Example two: Plaid and your bank

Plaid is the company that connects apps to your bank account. If you've ever linked your bank to Venmo, a budgeting app, or now an AI agent that helps with your finances, you've probably used Plaid without knowing it.

When you connect your bank through Plaid, Plaid creates what it calls an access token for that app. According to Plaid's own documentation, that token only works together with the specific app's credentials. It can't be used on its own, and it can't be handed to another app.

So if you switch from one AI agent to another, your bank connection doesn't come with you. The new agent has to go through Plaid again and get its own token. Plaid sits in the middle of every one of those connections. Agents come and go. Plaid stays.

That points to a clever two-sided advantage for players like Plaid and Linq. They don't have to spend money finding end users, because the startups they power do that for them, using the money those startups raise. Every new user an agent signs up becomes a new connection for Plaid or Linq, and they get that growth without paying for it. On the other side, those startups can't easily leave, because the tokens live with Plaid or Linq, not with the startup or the user. The relationship is held by the infrastructure layer, not the app. That's a double moat: free distribution on one side, lock-in that can't be moved on the other.

Example three: Composio and your work apps

Composio does for work apps what Plaid does for banks. It lets AI agents connect to tools like Slack, HubSpot, Gmail, and hundreds of others without each agent building those connections itself.

When you connect your Slack through an agent built on Composio, you sign in on a Composio page. Composio then stores the resulting credentials and uses them on the agent's behalf. Its documentation is direct about it: the app never receives your tokens, and Composio doesn't offer a way to export them.

That's good security. It's also a moat. The agent you're using doesn't hold the relationship with your Slack. Composio does. Composio is the "Plaid of app connectivity" for agents.

Why the app layer is losing

Compare what's portable and what isn't.

An AI assistant's memory can be rebuilt. Your preferences, your past conversations, your notes, all of it is just data. Export it, paste it into a new assistant, and it can learn about you again in an afternoon. The features can be copied. The interface can be copied.

But the token that ties you to your bank through Plaid can't be moved. The phone number thread that ties you to your assistant through Linq can't be moved. The Slack connection held by Composio can't be moved. Each of those has to be rebuilt from scratch, by you, every time you switch.

That's the pattern. The further you move away from the metal, the more replaceable you become. The closer you are, and the more you hold a session token that isn't portable, the harder you are to replace.

What this means for builders

If you're building at the app layer, assume your features will be copied. Your defense has to come from somewhere else.

If you're building in the middle, holding connections between users and the services they depend on, you're in the best position you've been in years. Every new agent that launches needs a way to text, a way to reach a bank, and a way to touch work apps. Each one becomes another customer.

The new moat isn't the app. It's the key. And the companies holding the keys are the ones that last.

No comments yet