The first 60 seconds with Ezra
The default failure mode of personal AI products is "tell me about yourself for twenty minutes." We did the opposite. Here's what your first minute with Ezra actually looks like.
Most products that need to "learn about you" do it wrong. They ask. A lot. A wizard, a survey, a checklist of preferences, a series of "what would you like Ezra to help with?" radio buttons.
That's setup. Setup is where products lose people. The longer the setup, the fewer people emerge from it still using the product.
So we made a rule for ourselves: setup happens invisibly while value is being delivered. The user is never "set up." They're just using it, and meanwhile, in the background, Ezra is figuring them out.
The actual flow
Sarah opens iMessage, taps her new contact "Ezra," types her first message. We're going to walk through the next sixty seconds, second by second.
T+0s — Sarah types: "Hi"
That's it. No form. No instructions. She just texts him.
T+1s — Ezra replies
Three things in the first reply: greeting, what he does, one question. He's not asking for permission. He's not selling. He's just starting a conversation.
T+15s — Sarah: "Sarah"
T+16s — Ezra
One question. He'll use her answer to set defaults — work-y vs. personal, professional vs. casual, what kind of help she actually needs.
T+25s — Sarah: "Real estate agent and a mom"
T+26s — Ezra
Three things going on here:
- "Got it — that's a lot" acknowledges the full picture (she has two demanding lanes)
- "Want to start with email?" suggests a single concrete first step (not a feature list)
- "I won't send anything without asking first" pre-addresses the biggest fear (AI doing things she didn't authorize)
All without making her configure anything.
T+30s — Sarah: "ok"
T+31s — Ezra sends a magic link
"Tap once to let me see your email" — not "authorize OAuth access to your Gmail account via the official Google permissions flow." Same thing, different language.
T+32-46s — Sarah taps. OAuth. Returns to iMessage.
While she's doing this, Ezra is reading her last few days of email in the background. Not all of it. Just enough to identify what's there.
T+47s — Ezra back, already reading
T+58s — Ezra
Specifics. A real name, a real listing, a real time, a real urgency assessment. She's not seeing a generic "I can help you with email!" She's seeing Ezra do email, on her actual inbox, before she's even confirmed she wants to use the product.
T+62s — She taps 👍, Ezra delivers a draft
The draft is in her voice — short, casual-professional, ends with "— Sarah" because that's how her last twenty sent emails ended. She didn't tell him to use that signature. He just inferred it.
What just happened
By the 60-second mark, Sarah has:
- An AI that knows her name, role, and basic context
- Gmail connected and being processed
- A specific, real, urgent thing identified from her actual inbox
- A draft reply in her voice ready to send
She has not:
- Filled out a form
- Configured preferences
- Picked from a feature list
- Read documentation
- Set up rules or filters
- Decided on a tone or style for Ezra
The setup happened. She didn't notice it. That's the goal.
The principle
Every product team thinks they're "minimizing setup." Most of them just shorten the wizard. We tried to remove the wizard entirely, and replace it with the conversation that was already going to happen anyway.
If you're going to text an AI assistant, you're going to tell it what you do. So Ezra asks once, naturally. If you're going to want it to read your email, you're going to authorize it. So Ezra suggests one tap, in the moment when it makes sense.
Setup, but in the shape of a conversation. Setup, but invisible.
If we got this right, when you're done with your first 60 seconds with Ezra, you don't think you "set up an AI assistant." You think you texted someone and got a draft back.