# How to build an agent that builds the culture — Zach Bruggeman, Ramp

## Executive summary

This talk details how Ramp leveraged its collaborative coding agent, Inspect, to fundamentally evolve its engineering culture. Beyond the technology itself, the focus is on lessons learned: using shared agent tooling to kill information silos, building trust through visible verification, and fostering collective innovation. Ramp reports that Inspect now writes and ships around 75% of the company's code.

## Key takeaways

- Killing Information Silos via Shared Tooling: By providing a single, standardized baseline (Inspect) across the organization, Ramp reduced the maintenance and setup cost associated with disparate individual setups, ensuring all teams benefit from the best practices.
- Building Trust Through Visible Verification: The agentic process is made transparent. When a pull request is reviewed, engineers can see the exact conversation, the tools used, and even video recordings of the verification agent proving the change holds up, making it effortless to trust the outcome.
- Fostering Collective Innovation: The culture shifted from 'lone wolf' operation to collaboration. Engineers spontaneously built new agents, such as Review Buddy, by leveraging existing automation triggers within Inspect, leading to rapid, self-driven improvements.

## Technical details

- Inspect Agent Capabilities: Inspect is a fully multiplayer, collaborative coding agent that utilizes built-in design primitives, review agents, and verification flows. It is connected company-wide to critical data sources like telemetry and databases.
- Agentic Debugging Example: A telemetry bug was publicly investigated using Inspect. The issue was identified as a Cloudflare platform limitation where the library calls fetch but never reads or cancels the response body, which Inspect successfully patched.
- Review Buddy Implementation: The Review Buddy agent was developed by internal engineers who utilized Inspect's existing automation triggers (e.g., when a pull request was opened) to build a specialized, company-wide code review tool.

## Practical implications

- Organizations can standardize their AI tooling baseline to reduce knowledge disparity and setup overhead.
- Implementing transparent, auditable AI workflows (e.g., showing agent conversations and verification steps) is crucial for building team trust.
- Encourage engineers to treat AI tools not just as individual productivity boosters, but as platforms for collective, self-driven innovation.

## Topics

AI Engineering, DevOps Culture, Agentic Workflow, Code Review, Information Silos, Inspect, Zach Bruggeman

Source: https://www.youtube.com/watch?v=MDI81OaZAD4
