---
title: "How Ramp Uses Langfuse to Build AI Agents That Improve Themselves"
seoTitle: "How Ramp Uses Langfuse to Build Self-Improving AI Agents"
date: September 18, 2026
description: "Ramp's engineering team on tracing agents with self-hosted Langfuse, and how Reflect turns traces into merged fixes - 30% faster sessions, 15% fewer tool calls."
ogImage:
tag: customer-story
customerLogo: "/images/customers/ramp/ramp-light.svg"
customerLogoDark: "/images/customers/ramp/ramp-dark.svg"
customerQuote: "We wanted something built for agents as users first. And that means API first. An agent should never get stuck waiting for a human because of a deficiency in the API. This is what Langfuse is."
customerQuoteHighlight: "API first"
customerQuoteTag: "Video story"
quoteAuthor: "David Traina"
quoteRole: "Data Platform"
quoteCompany: "Ramp"
ctaLabel: "Watch"
showInCustomerIndex: true
---

[Customer stories](/users) › Ramp

<iframe
  width="100%"
  src="https://www.youtube-nocookie.com/embed/2P6xL6eVTPo"
  title="How Ramp uses Langfuse: Observing, debugging, and continuously improving production AI agents"
  frameBorder="0"
  allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share"
  referrerPolicy="strict-origin-when-cross-origin"
  className="aspect-video w-full rounded border mt-2"
  allowFullScreen
></iframe>

<p className="text-sm text-primary/60 mt-2 mb-8">
  <a href="https://www.youtube.com/watch?v=2P6xL6eVTPo">Watch on YouTube</a>
</p>

## Highlights from the talk [#highlights-from-the-talk]

**AI instrumentation**

![Langfuse trace view for Ramp's durable_agent/ap_coworker_agent showing nested turns, tool calls, and generation metrics](/images/customers/ramp/instrumentation-trace.jpg)

**Why Langfuse**

![Why Langfuse slide highlighting API-first design with an agent querying 600M observations across Ramp Langfuse projects](/images/customers/ramp/why-langfuse-api-first.jpg)

**Working with the team**

![Slack threads showing Ramp collaborating with the Langfuse team on features, SDK fixes, and session view feedback](/images/customers/ramp/working-with-langfuse-team.jpg)

**Ramp Reflect**

![Ramp Reflect workflow diagram from production sessions through Langfuse enrichment to deeper reasoning and human review](/images/customers/ramp/reflect-workflow.jpg)

**Reflect's impact**

![Reflect impact on Ramp Research showing a before and after fix path with 31% faster sessions and fewer tool calls and tokens](/images/customers/ramp/reflect-impact.jpg)

[Ramp](https://ramp.com/) is a smart financial infrastructure company serving 70,000+ businesses across cards, expense management, AP, travel, treasury, procurement, and token spend management - saving customers roughly $2B and 20 million hours. Nearly all of it now runs on agents.

In this talk, Ryan Delgado (Engineering) and David Traina (Data Platform) walk through Ramp's agent portfolio and the observability layer underneath it. Customer-facing agents include Policy Agent, which approves or denies expenses before a manager ever sees them, and Stack, which compresses month-end close from a week and a half to 1-3 days. Internally, Inspect - Ramp's in-house background coding agent - now accounts for ~70% of merged PRs at roughly 1,500 PRs per day, and Ramp Research answers data questions in Slack in 2-3 minutes instead of 3-5 hours.

> “Inspect accounts for about 70% of the merged PRs within Ramp, and we're raising about 1,500 PRs a day across a few different repos.”
>
> — Ryan Delgado, Engineering, Ramp

David Traina explains why Ramp chose Langfuse: fully [self-hosted](/self-hosting) on standard infra (S3, Redis, Postgres, ECS, Terraform) with ClickHouse Cloud, fully open source, [OpenTelemetry](/integrations/native/opentelemetry)-native, and API-first so agents - not just humans - can be first-class users.

He then demos Reflect, a Ramp agent that monitors other agents. Reflect scores and clusters Langfuse [traces](/docs/observability/overview), diagnoses recurring failure patterns, and proposes fixes. Applied to 12 agents with 120+ merged suggestions, it cut Ramp Research token usage by 10-20%, tool calls by 15%, and session times by 30%.

> “Reflect works by semantically scoring those traces and grouping similar traces together… it analyzes those traces in aggregate to find patterns, diagnose issues, and propose fixes for a human owner to either accept or reject. We've seen a 10-20% decrease in token usage, 15% fewer tool calls, and 30% faster end-to-end session times.”
>
> — David Traina, Data Platform, Ramp

## Ready to get started with Langfuse?

Join thousands of teams building better LLM applications with Langfuse's open-source observability platform.

- [Start free](/cloud)
- [Documentation](/docs)
- [Talk to an expert](/talk-to-us)

<!-- agent-instructions -->

---

## Agent Instructions

This page is part of the [Langfuse](https://langfuse.com) documentation, published as plain Markdown for AI agents. Every page is available as Markdown by appending `.md` to its URL, or by sending an `Accept: text/markdown` header. This page: `https://langfuse.com/users/ramp.md`.

### Querying these docs

If the answer is not on this page, query the documentation instead of guessing:

- **Semantic search** across all Langfuse docs, returning an answer with the relevant pages and excerpts. Ask a specific, self-contained question:

  ```bash
  curl -sG "https://langfuse.com/api/search-docs" --data-urlencode "query=How do I trace a LangGraph agent?"
  ```

- **Index of every page**: <https://langfuse.com/llms.txt>, with per-section indexes [llms-docs.txt](https://langfuse.com/llms-docs.txt), [llms-integrations.txt](https://langfuse.com/llms-integrations.txt), and [llms-self-hosting.txt](https://langfuse.com/llms-self-hosting.txt).

### Before writing Langfuse code

- **Install the [Langfuse Agent Skill](https://langfuse.com/docs/api-and-data-platform/features/agent-skill).** It encodes Langfuse's own best practices for instrumentation, prompt management, and evaluation, and materially improves results.
- **Read [What does a good trace look like?](https://langfuse.com/docs/observability/best-practices.md)** before instrumenting an application.
- **Verify endpoints, parameters, and response fields** against the [API reference](https://api.reference.langfuse.com) instead of inferring them from code examples.
- **Use the [Langfuse CLI](https://langfuse.com/docs/api-and-data-platform/features/cli)** (`npx @langfuse/cli api <resource> <action>`) to read or write traces, prompts, datasets, and scores from the terminal.

Found an error in these docs? Please open an issue at <https://github.com/langfuse/langfuse-docs/issues>.