Why We Built Our Own Background Agent
Source: https://builders.ramp.com/post/why-we-built-our-background-agent Author: Zach Bruggeman, Jason Quense, Rahul Sengottuvelu (Ramp) Date: 2026-01-13
Summary
Ramp’s engineering team explains why they built a custom background coding agent (Inspect) instead of using existing solutions. Inspect runs autonomously: tests code, reviews data, monitors systems — without needing a human in the loop.
Key Claims
- Why custom: existing agents (Claude Code, Devin) are interactive — optimized for human-in-the-loop. Ramp needed true background operation.
- Inspect’s design: asynchronous, event-driven. Triggered by CI failures, data anomalies, or scheduled tasks — not by human prompts.
- Key capability: can run for hours unattended; checkpoints progress; resumes after interruption.
- Success metric: time saved for engineers. Inspect handles 30-40% of routine debugging tasks at Ramp.
- Biggest challenge: reliability at the tails. The 5% of cases where Inspect makes an error are more costly than the 95% it handles well.
Entities
- Ramp — fintech/expense management company
Concepts
- Coding Agents — background agent design pattern
- Autonomous Research — autonomous debugging = research on codebase
- Context Engineering — event-driven context injection for background agents