Three months ago, every startup I talked to wanted background agents. Fire off a task, let it run in the cloud, review the result later. Now those same teams are asking for something bigger. They want all of their development in the cloud, synchronous sessions included. Not just the async work, everything. The local machine is becoming a thin client.
The infrastructure world is scrambling to catch up. The sandboxing providers built for ephemeral, fire-and-forget containers are realizing that synchronous coding in the cloud demands real VMs, persistent, fast, indistinguishable from a local machine. When a developer is sitting inside the session, latency and fidelity stop being nice-to-haves. The bar is 60 frames per second and a terminal that feels native, because the comparison is the laptop in front of them.
Here is what surprised us building for this shift. Synchronous and asynchronous cloud agents turn out to be almost the same thing. The VM, the environment, the agent loop, the session tracking, all identical. The only real difference is whether a human is present while it runs. A platform built for background execution is one interface away from serving the fully synchronous workload, which is exactly why cloud dev environments are the unlock for parallel agent work. Get the substrate right and both modes fall out of it.
Meanwhile, most enterprises are roughly six months behind this frontier, and pretending otherwise is how you build a product nobody buys. What enterprises want right now is concrete. Tag an agent in chat, get a pull request, run the whole thing inside their VPC. That workflow has to be excellent today, because that is where the budgets are. But the startups on the frontier are a preview of where those enterprises will be in two quarters.
The strategic play is to serve both ends of that timeline with one architecture. Build for where the frontier already is, package it for where the enterprise actually is, and let the market walk toward you. The gap between the two is not a problem. It is the roadmap.
Key takeaways
- The frontier has moved from async background tasks to running all development sessions, sync and async, in the cloud.
- Architecturally, synchronous and asynchronous cloud agents are nearly identical, and platforms built for one can serve both.
- Enterprises run about six months behind the frontier, so the winning product serves today's enterprise workflow while being positioned for where startups already are.
FAQ
What changed to make synchronous cloud development viable?
Sandbox infrastructure matured from ephemeral containers to real, persistent, low-latency VMs, and coding agents got reliable enough to run continuously in the cloud. Together those remove the tradeoffs that made local development feel necessary.
Should enterprises jump straight to fully cloud-based development?
Most are not there yet, and that is fine. The practical entry point is event-driven work, like tagging an agent in chat and getting a reviewable pull request. The important thing is choosing infrastructure that supports that today and full cloud sessions later, so the migration is a setting change rather than a replatform.
Related Essays
Cloud Dev Environments Are the Unlock for Parallel Agent Work
Running agents locally hits a wall fast. Cloud-based dev environments let teams run parallel agent sessions, share work instantly, and actually scale.
Every CTO Is Asking How to Move Engineering Work to the Cloud
Startups are moving all of their engineering work into cloud agent environments. Enterprises are moving some. The constraint that decides how is the codebase itself.
Democratizing Software Engineering
AI coding is not about replacing developers. It is about expanding who can request software work in the first place — and 10x-ing total demand.
Drowning in pull requests that need your review? Try Tembo Review, a beautiful AI-assisted PR review tool unlike anything you’ve used.