Why your AI assistant should live in SMS, not in another app

An AI assistant should live in SMS because SMS is the most universal interface on Earth. It's already on every phone, works across every carrier, requires no download, and has zero learning curve. Putting an AI agent in SMS removes every friction between you and the assistant — no app to download, no login to remember, no UI to learn. You save a number in your contacts. You text it. It works.

The standard playbook for an AI startup looks like this: build an app, drive installs, retain inside your app, monetize inside your app, panic when the platform owner ships a competing feature. We've watched this story end the same way for fifteen years.

SMS breaks the playbook in a useful way.

What you give up by being an app

Apps look like the win because they give you control of the surface — the chrome, the navigation, the notifications, the gradient hero. But the cost of all that control is a stack of frictions you've stopped noticing because every product you use has them:

• Install friction. Some percentage of people who hear about you will never download anything. That percentage is large and growing.
• Account friction. Email, password, verification, password reset on the second visit because nobody remembers it.
• Permission friction. Push notifications, contacts, microphone, camera. Each prompt costs you a percentage of users.
• Update friction. Your latest feature only works for users on a recent OS, on a recent app version, who happened to update.
• Attention friction. Your app is one icon among 200. The user has to remember it exists, find it, open it, and re-orient. Every time.

An SMS agent has none of these. There's no install. No account. No permissions request. No update cycle. The "icon" is the contact card the user already saved. The "open" is the same thumb motion they use to text their spouse.

What you gain by living in SMS

Beyond the negative case (what you don't have to overcome), there are real positive properties of SMS as an agent surface:

It's the only interface that's truly universal. Every carrier, every device, every operating system, every age, every income bracket. iMessage upgrades the channel when both ends support it. RCS does the same. The fallback is plain SMS, which has a 100% success rate against any phone made in the last twenty-five years.

It's the highest-attention channel. A text message is read within minutes by 90%+ of recipients. Email open rates above 30% are aspirational. App push notifications are mostly ignored. Texts are not.

It enforces simplicity. A 160-character constraint forces the product to be conversational. You can't hide behind UI. The agent has to actually be useful in plain language. That's a feature, not a limitation.

It composes well with everything else. An SMS agent can send a link to a richer view when the moment requires it — a calendar pick, a Stripe checkout, a doc preview. The text channel is the front door; richer interactions live behind links, summoned only when the situation calls for them.

The objections

"What about security?" SMS itself is not end-to-end encrypted, but iMessage and RCS upgrade the channel automatically when both ends support them. For 99% of personal AI use cases, the carrier-level transport security is adequate. A well-designed SMS agent never asks for sensitive credentials over the channel itself — payment, identity verification, sensitive data all happen behind links to authenticated views.

"What about features?" You lose visual chrome. You don't lose substance. Most of what an agent does — remembering, drafting, scheduling, taking action, talking to other agents — is text-native. The richness lives in the agent's reasoning, not in dropdown menus.

"What about voice?" Voice is a separate channel and a great one. An SMS-first agent can absolutely add voice as a parallel input — and the same agent identity works seamlessly across both, because the memory and identity layer are separate from the input/output channel.

The deeper argument

Apps were the answer to "how do we deliver software on a phone?" Agents are the answer to "how do we deliver intelligence anywhere?" The interfaces are different because the products are different.

An app is a place you visit to get a thing done. An agent is a relationship you maintain because it's quietly working on your behalf. The right interface for visiting a place is a destination — an app you open. The right interface for maintaining a relationship is a conversation — a thread you return to.

Phones already have a thread. Every contact you've ever texted is one. The agent should live in there with the rest of the people who matter to you, because in some real sense it is one of them now.

Where this leaves us

SMS isn't the right channel for every product. It is the right channel for a personal AI agent. Apps will continue to exist, and there are plenty of agents that live productively inside specialized apps. But the "one agent that knows me, lives with me, and shows up wherever I am" pattern fits SMS better than it fits anything else — at least until something better comes along.

And nothing better has come along. Yet.