The work is still there when the agent comes back
A coding agent can open a pull request in the afternoon and wait until the next morning for a review. When the reviewer asks for a change, the agent needs the same checkout, installed packages and files it was already working with. [1][3]
Superserve gives an agent its own isolated virtual machine in the cloud. When the agent has to wait, Superserve can pause the machine and save its memory, running processes and disk. The agent can resume the same machine for the next turn, or fork a snapshot to try another approach. [1][2]
Phaser gives each game its own workshop
A Phaser Game Agent user describes a game and gets back a playable link. Behind it, Phaser starts a Superserve sandbox with its game engine, build tools and a browser already installed. The agent builds the game and tests it in that environment. [4]
Players can try the published game while the sandbox is paused. When they ask for a change, the agent can resume the saved workspace, edit the game and publish the next version. Superserve stops charging for compute while the machine is paused. [2][4]
From coding agents to YC’s QM
Pavitra Bhalla ran into the infrastructure problem while building code reasoning agents at Dosu. Each agent session needed a place to work with a codebase, a shell and its tools. He started Superserve to make that environment easier for other teams to provide. Superserve joined Y Combinator’s Fall 2026 batch. [5]
Bhalla told SOTA2 that Y Combinator runs its own instance of QM on Superserve. QM is YC’s open-source agent harness for teams. It gives people and shared project rooms their own workspaces, where an agent can return to the files and context it was using. [5][6][7]
Where Superserve ranks
The bigger picture
Starting quickly matters when an agent needs a fresh machine. Keeping the work matters when the next instruction comes an hour later. ComputeSDK and AI Sandbox Bench publish separate measurements for stages such as startup and resume, so teams can check the parts of the workflow they use most. [8][9]
Written by SOTA2 from public documentation, published customer examples and information Superserve’s founder provided. The SOTA2 rankings are separate from the benchmarks linked above.
Sources
- Superserve · Virtual machines for long-running agents
- Superserve Docs · Pause, resume and delete
- Superserve · Cursor Cloud Agents integration
- Superserve · Phaser Game Agent case study
- Y Combinator · Superserve company and launch profile
- Y Combinator · QM open-source project
- Superserve founder input · Y Combinator’s QM deployment
- ComputeSDK · Sandbox provider ratings
- AI Sandbox Bench · Cloud sandbox measurements
