Jesse Johnston

I’m a product leader in multi-device computing and a builder of AI systems. This site is where I write down how those systems actually work, including the parts that broke.

What I build

At Google, I lead product for Android’s multi-device ecosystem, the work of making phones, tablets, and laptops behave like parts of one intelligent system for more than a billion people. My work spans the common foundation underneath those experiences and the ambient, AI-native experiences built on top of it.

I have been building longer than I have been a PM. I started a computer-services business as a teenager and ran it for eight years, which paid my way through college. Since then I have taken products from 0→1 into scaled platforms, led product teams, and earned patents across connectivity, UX, and system design.

At home, the users share my last name and the infrastructure lives in the garage. I run the family cloud, automate the house, host local models, and build agents that sort email, investigate alerts, plan workouts, and maintain the systems underneath all of it. These are personal projects, but not demos. They touch real mail, real devices, and services my family depends on, so they have to earn authority, expose evidence, and fail safely.

The Agents page says what those systems run as today. Patterns collects the rules they have earned.

The same problem at two scales

I keep coming back to continuity: context should survive the boundaries between devices, tools, and sessions.

A phone should be able to help a tablet recover. A workout prescription should move from an agent, through my approval, onto my watch, then return as performance data. A coding session should still be waiting after I leave my desk and attach from a parking lot.

At work, the same problem operates at a much larger scale. The user should not have to reconstruct the task every time the screen changes.

How I build

Most of my projects now start with a command-line AI agent working inside the actual repository, not a chat tab offering advice from the sidelines. The model can read files, run tools, make a branch, and show me the diff. It can draft and propose; it cannot make the work mine. I review it, test it against the real system, and answer for what ships.

The terminal in my pocket explains the cockpit that lets the work follow me away from my desk. An employee handbook for my AI staff explains how each new session inherits the context and decisions that came before it. Building a blog I’d actually keep up is the colophon for the site itself, including the AI-assisted drafting process and the voice specification behind it.

Selected work

• Eight pull requests to my manager’s repo

  An AI agent produced eight stacked pull requests for a real Android app; I reviewed, tested, and signed the work for the manager who also reviews me.

• The terminal in my pocket

  The old plumbing that lets an AI job keep running after I leave my desk and lets me resume it from my phone.

• The guardrail Gmail wouldn’t sell me

  The boundary around an email agent: what it may do, what it cannot do, and why the boundary had to be plain code.

Connect

The simplest way to reach me is email. I’m also on LinkedIn, GitHub, and RSS.