Most of my best engineering ideas never happen while I am sitting at my desk staring into a split code editor.

They usually strike right in the middle of everyday life: walking down the quiet sidewalk after dropping my daughter off at school, waiting by the produce aisle at the grocery store, or strolling around the block on a crisp autumn afternoon. In those moments, your mind is uncluttered, and suddenly a solution to a stubborn architectural bug or a new tool idea just snaps into place.

For a long time, this was accompanied by a familiar frustration. I would try to jot down a frantic note on my phone, but by the time I made it back home, took off my boots, unpacked the groceries, and sat back down in front of my monitor, that delicate spark had either cooled down or gotten buried under household chores.

Google Antigravity 2 is remarkably capable. Once you set a clear goal, it can research repositories, refactor modules, and run full test suites autonomously. But it suffered from one obvious constraint: it lived exclusively on my local desktop machine. If I wanted to run a task or review its progress, I had to physically remain in my chair.

I didn’t want to spend more time glued to a desk. What I really needed was a way to let my assistant work for me while I was living my life outside.

Speaking Ideas on the Sidewalk

That was the seed that grew into Antigravity Pilot (agy-pilot).

When thinking about how to bridge my local machine to my pocket, I considered building a quick web app or hooking up a Discord bot. But neither felt right for real life on the move. Typing code snippets or long prompts on a cramped virtual keyboard while walking down the street is clumsy and frustrating.

Telegram, on the other hand, offered two distinct superpowers that made it the perfect fit.

First, its interactive inline keyboards allowed me to build clean, nested menus right inside the chat window. With a couple of taps, I could easily switch between active workspaces, browse ongoing conversation threads, and approve high-privilege tool requests without ever refreshing a screen. Because it communicates via long polling, it works effortlessly behind home routers without messy port forwarding or complex tunnels.

Second—and easily my favorite part—is native voice messaging.

Now, when an idea hits me while walking back from the store with a bag of groceries under my arm, I don’t stop on the pavement to peck at a screen. I simply hold down the microphone button, talk through what I want to test or investigate, and let the bot transcribe and forward my instructions directly to Antigravity back at home.

Standing on the Shoulders of Giants

In many ways, this approach shares the same local-first spirit as tools like OpenClaw, which connects private AI agents to everyday messaging apps. But rather than attempting to reinvent another general agent framework, I chose to stand squarely on the shoulders of giants.

Antigravity already has deep reasoning, comprehensive workspace awareness, and robust execution tooling. All it needed was a reliable companion daemon—a lightweight, non-intrusive bridge that speaks its internal language server protocols and translates them into an intuitive mobile interface.

Of course, getting there came with its share of engineering hurdles.

One immediate puzzle was handling restarts: if my Mac rebooted or Antigravity’s core process restarted, the bridge had to smoothly discover the newly assigned hub port and authentication token without manual intervention.

Another subtle challenge was conversation step flooding. Antigravity 2 frequently generates twenty or thirty distinct execution steps for a single prompt—inspecting files, running commands, and evaluating diffs. If you push a new chat notification for every step, your phone will buzz nonstop and bury your chat in noise. I had to design a progressive update mechanism that inlines and collapses intermediate traces, keeping the conversation stream clean while leaving every detailed log accessible behind a single tap.

The Quiet Satisfaction of Small Bridges

I am currently working on making the bridge invoke specialized skills even more naturally on demand. But even in its current form, it has completely changed how I think about my daily development workflow.

Just yesterday afternoon, I sent a voice note describing a refactoring idea for a utility library while picking up a box of fresh apples at the local market. Ten minutes later, while waiting at a pedestrian crosswalk under the golden evening sun, my phone buzzed with a short notification. I looked down: Antigravity had reviewed the files, completed the changes, and verified that all ninety unit tests had passed.

It wasn’t some grand AI revolution. It was just a small, practical bridge that gave me back an hour of freedom to walk in the fresh air without losing a good idea. And at the end of the day, that is the kind of software craftsmanship I care about most.

If you’re curious about the mechanics or want to poke around under the hood, the project is open source on GitHub: donileong/agy-pilot.