Eve’s Slack-driven background agent WALL-E authors 22% of merge requests in its monorepo, 53% of them one-shot
Eve, a legal technology company, built WALL-E, a background coding agent engineers call from Slack. Each Slack thread gets an isolated Modal sandbox with a full copy of the application, real database, backend and frontend; WALL-E runs OpenCode inside it, tests in a real browser, checks data in the database, and posts a merge request with a live sandbox link, typically in about 15 minutes. It reaches Linear, Sentry, Figma and Playwright through MCP. Before a sandbox shuts down, WALL-E runs a retrospective and commits reusable lessons to the nearest AGENTS.md. It began as a hackathon project, opened its first real merge request two weeks later, and now authors 22% of merge requests in the monorepo.
The numbers (as reported)
| Share of merge requests | 22% of all merge requests in the monorepo |
|---|---|
| One-shot rate | 53% of WALL-E merge requests need no human iteration |
| Time per merge request | About 15 minutes |
| Hackathon to first real merge request | Two weeks |
Facts
- Company / subject
- Eve (Eve Legal)
- Tools
- OpenCode
- In-house tools
- WALL-E
- Codebase
- brownfield
- Work type
- Company rollout
- Stack
- Modal, MCP, Playwright
- Published
- 2026-03-19
- Last checked
- 2026-11-13
What the source claims Practitioner (own blog/paper)
Claimed by the builder, writing about their own work. Not checked independently.
- WALL-E authors 22% of all merge requests in the monorepo 22%
- 53% of its merge requests are one-shot, with no human iteration 53%
“53% of those are one-shot”
- Merge request ready in about 15 minutes about 15 minutes
Caveats
Written by Eve’s own engineer. The post gives no absolute counts of merge requests, team size, cost or time period, and does not name the models used. One-shot means no human iteration before the merge request, as defined by Eve.
More stories by tool: OpenCode
Related topics: Company rollouts and costBrownfield codebases