Development Is Leaving Your Laptop

Remote development environments for agents are becoming standard across coding agent platforms. Anthropic has cloud VMs for Claude Code, Amp has Orbs, and OpenCode thinks about the same idea from time to time. Infra is also evolving to make it easy to start and stop VMs and keep somewhat persistent environments.

My only real use of remote VMs has been Orbs in Amp. It gave me a “feel the future” feeling: using Amp from my laptop, then my phone, continuing throughout the day on whatever device I had. It was cool and lasted a few days.

Lately I’ve been doing a bit of iOS development. This has to be done locally and requires Xcode to compile the code and preview what I’m building. I can start a session on my laptop and continue on the web or phone, but my laptop has to be awake for this. That’s a big limitation. I also need to plug my phone into my laptop or be on the same WiFi to test what I’m building.

Besides this, dev work can run on VMs that most users never need to think about. This makes parallel work easy. For small tasks it’s ideal. Fixing bugs and building small features is trivial, which means a lot of work can be automated, leaving the more interesting work to humans. Anything that requires problem solving, or is interesting at all, needs my full focus. Working on fewer, more important tasks is how I do anything of value. Still, ephemeral environments are a useful progression. Whatever I’m working on, I see myself working with an agent that does most of the grunt work and some thinking.

The progress so far is mostly for the trivial work I mentioned. A few missing pieces need to come next.

Local dev usually means running the project: one or more servers, or one or more containers, exposing a website or API for me to use. Moving this to remote VMs introduces challenges. The VMs need to be persistent. Some companies start and stop them based on use, which makes sense since they sit idle most of the time and saving money matters. But I should have an easy way to say “hey, I need this active now for testing”, or to let someone else test my work. Which brings me to the next point.

Environments need a secure way to access the services, potentially fronted by an auth service like Cloudflare Access. I’d recommend exe.dev here; they’ve made this feel like magic. Running your dev environment remotely also makes sharing and reviewing PRs straightforward. I have yet to see a coding agent platform do this well. Amp gets its agent to share a screenshot of the website in the conversation. That’s okay, but I care more about UX than UI.

What about data? Most products need it to test thoroughly. AI makes seed data easy to create, but getting it into remote environments now matters more. Maybe this isn’t a problem at all, and database branching (e.g. Supabase, PlanetScale, Neon) becomes widespread. A separate database is essential, given that AI can still wipe your entire production database.

These problems will be solved sooner rather than later, and probably are being solved right now, either ad-hoc by individuals or as product features.

For now, I’d delegate simple tasks to a remote coding agent, and use a persistent dev box for complex ones. Agents can run for a long time, so it’s nice to close my laptop, travel somewhere, and pick up where I left off without interrupting them.