AIDevOps

How Ramp Built Self-Maintaining Software

Data Driven NYC
Data Driven NYC•May 5, 2026

Why It Matters

Self‑maintaining, AI‑powered pipelines dramatically cut engineering overhead while improving reliability, paving the way for fully automated software factories that can evolve products without constant human intervention.

Key Takeaways

  • •Ramp Labs built Inspect, a sandboxed AI coding agent for maintenance.
  • •Nightly stateless runs surfaced bugs but suffered diminishing returns and noise.
  • •Persistent monitors with state reduced false alerts and enabled targeted fixes.
  • •Signal‑to‑noise ratio proved critical; triage patterns filtered noisy or duplicate alerts.
  • •Self‑maintaining pipeline hints at future AI‑driven software factories and product improvements.

Summary

Alex from Ramp Labs explains how the team created a self‑maintaining software pipeline for Ramp Sheets using an internal AI agent called Inspect. Inspect runs code in isolated sandboxes, integrates with GitHub, Datadog, Sentry and other tools, and can automatically generate pull requests to fix detected issues.

The initial approach was a nightly, stateless cron job that scanned for security problems, regression failures and latent bugs. While it caught real defects, the lack of state caused diminishing returns and overwhelming noise from repetitive alerts and massive telemetry data. To address this, the team introduced persistent monitors tied to specific code paths; when a monitor fires, Inspect evaluates its merit, updates its description, and either creates a fix or silences duplicate alerts.

A concrete example involved a loophole that let a competitor benchmark Ramp Sheets despite an email ban. Inspect flagged the recurring login issue, generated a PR, and the team merged a comprehensive fix. The speaker emphasizes that high signal‑to‑noise is essential—noisy monitors erode trust—and that fine‑grained observability drives empathy for user pain points.

The broader implication is a shift from manual code upkeep to AI‑driven software factories that not only maintain but also iteratively improve products. By chaining Inspect sessions and leveraging dynamic monitoring, organizations can scale maintenance, reduce human toil, and eventually let agents propose feature enhancements based on real‑world usage data.

Original Description

In this talk from Data Driven NYC, Alex Levinson, Software Engineer at Ramp Labs, dives into the future of "Self-Maintaining Software." As AI takes over the heavy lifting of writing code, the new bottleneck in the development lifecycle is maintenance and observability. Alex introduces Ramp Inspect, an agentic system designed to autonomously monitor production, triage alerts, and propose code fixes without human intervention. From building a "nightly QA pass" to the visionary concept of the "Software Factory," learn how Ramp is shifting the engineering focus from building software to building the pipelines that build—and maintain—software at scale.
00:04 - Introduction to Ramp Labs and self-maintaining software
01:11 - The shift from writing code to code maintenance
01:59 - Introducing Ramp Inspect, the background coding agent
03:05 - The first experiment: Nightly AI code automation
04:23 - The limits of stateless monitoring in large observability surfaces
05:47 - Using Datadog monitors to give the AI state and focus
07:23 - Real-world example: AI autonomously fixing an authentication bug
08:14 - How to control noise and implement an AI triage pattern
09:27 - The old vs. new paradigm for continuous code observability
10:21 - Key learnings on building autonomous AI software factories
Ramp Labs
HOSTED BY:
FirstMark Capital
Matt Turck (Managing Director)
This session was recorded live at a recent Data Driven NYC, our in-person, monthly event series. If you are ever in New York, you can join the upcoming events by following FirstMark on Luma: https://luma.com/firstmarkcap
Check out the MAD Podcast:

Comments

Want to join the conversation?

Loading comments...