Quality assurance · Production

Ramp builds a full-context background coding agent on Modal that writes over half of all merged pull requests

The problem

Ramp wanted to give every builder—engineers, product managers, and designers—access to AI-assisted coding, but local agents required dev environment setup, couldn't integrate with internal tooling, and any background agent slower or less capable than local would not get adopted.

Workflow diagram · grounded in source
1
Cron filesystem snapshot build
integration
“Every 30 minutes, a Modal Cron clones each of Ramp's repositories, installs all dependencies, runs initial builds, and saves a snapshot of the Sandbox's filesystem”
2
Builder sends prompt
trigger
“Queues route prompts from any client — Slack, web, Chrome extension — into the right session”
3
Sandbox spun up from snapshot
integration
“Inspect creates a new Sandbox from the latest snapshot. Since the snapshot is at most 30 minutes old, syncing with the head of the repo is nearly instant. End-to-end, sessions start working on a prompt in a few seconds.”
4
OpenCode coding agent runs
ai_action
“Inside the sandbox, OpenCode runs as the coding agent, alongside a VS Code server for manual edits, a web terminal, and a VNC stack with Chromium for visual verification of frontend changes”
5
Visual verification by agent
validation
“The agent can take before-and-after screenshots, navigate the app in a real browser, and confirm its work visually—just as a human would”
6
Review-ready pull request
output
“The current generation is already producing review-ready pull requests”
Reported outcome

Within a couple of months, roughly half of all merged pull requests at Ramp are started by Inspect, and over 80% of Inspect itself is now written by Inspect, with product managers and designers empowered to ship code directly.

Reported metrics
merged pull requests started by Inspect (intro)over half
merged pull requests started by Inspect (impact section)roughly half
Inspect code written by Inspectover 80%
Session startup timea few seconds
Show all 5 reported metrics
merged pull requests started by Inspect (intro)over half
merged pull requests started by Inspect (impact section)roughly half
Inspect code written by Inspectover 80%
session startup timea few seconds
concurrent sessions supportedhundreds of concurrent sessions
Reported stack
ModalModal SandboxesOpenCodePostgresRedisTemporalRabbitMQGitHubSlackBuildkiteSentryDatadogLaunchDarklyChromium
Source
https://modal.com/blog/how-ramp-built-a-full-context-background-coding-agent-on-modal
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Within a couple of months, roughly half of all merged pull requests at Ramp are started by Inspect, and over 80% of Inspect itself is now written by Inspect, with product managers and designers empowered to ship code…

What tools did this team use?

Modal, Modal Sandboxes, OpenCode, Postgres, Redis, Temporal, RabbitMQ, GitHub, Slack, Buildkite.

What results were reported?

merged pull requests started by Inspect (intro): over half; merged pull requests started by Inspect (impact section): roughly half; Inspect code written by Inspect: over 80%; Session startup time: a few seconds (source-reported, not independently verified).

How is this quality assurance AI workflow structured?

Cron filesystem snapshot build → Builder sends prompt → Sandbox spun up from snapshot → OpenCode coding agent runs → Visual verification by agent → Review-ready pull request.

Related quality assurance cases