The Pragmatic Engineer · Video summary

Ramp: Lessons from Building a New AI Product - The Pragmatic Summit

Ramp shares lessons on building AI products, emphasizing a single agent with many skills, iterative development, and a culture shift for faster, impactful delivery.

The Pragmatic Engineer · March 9, 2026 · 37-min episode · 6-min read

Ramp: Lessons from Building a New AI Product - The Pragmatic Summit

Key takeaways

🎯Shift from building thousands of specialized agents to a single agent with a thousand skills to drive your AI framework.04:10
🚀Consolidate disparate AI agents into a single, versatile agent with a broad skill set to streamline development and management.04:10
💡In the new AI paradigm, software should focus on building autonomous systems of action that can react, reason, and act with minimal human supervision.05:01
🚀Consolidate conversational interactions into a single conversational UX (Omnihat) deployed across all product surfaces, complementing traditional UIs.05:15
🔍AI products cannot be oneshotted; start simple and iterate, ensuring alignment across product, design, and engineering that perfection isn't immediate.10:54
📈The context provided to LLMs is more critical than the models themselves; learn from live data to extract relevant context like employee roles and titles.11:41

Chapters

0:05🎯 Introduction to RAMP and AI at the Summit
0:15🎯 Simplified Expense Use Case & Time Savings
0:29🎯 Building Automated Agents and Paradigm Shift
6:16🎯 Consolidating Interactions and Building a Framework
7:11🎯 Policy Agent and Real-Time Spend Classification
10:10🎯 Rethinking Rules Using Natural Language
11:53🎯 Learning and Iterating with Internal Data
16:28🎯 Applying Data Labels and Evals for Confidence
18:47🎯 Integrating Cybersecurity and Versioning
23:34🎯 Cultural and Technical Principles for AI Infrastructure
26:09🎯 Handling Hallucination, Safety, and Trust
26:40🎯 Off-the-shelf Tool Catalog and Ecosystem
28:06🎯 Internal Context Integration and Ramp Inspect
31:19🎯 Impact of Ramp Inspect on Development Workflow
32:50🎯 Team Divergence and the Value of Judgment
36:32🎯 Reflections on Building AI-Driven Finance Products

The digest

Ramp, a leading finance platform serving over 50,000 businesses, shares insights from creating AI-powered agents that automate expense management and finance workflows. This detailed briefing covers their transition from numerous specialized agents toward a unified conversational AI, the design and iterative development of their policy enforcement agent, and the complex infrastructure and cultural shifts required to embed AI deeply into product and engineering processes. These lessons provide a roadmap for companies aiming to harness AI to reduce manual finance work and accelerate product delivery with safety and scalability.

High-Level Overview

Ramp is transforming finance operations for modern businesses by automating time-consuming manual tasks, starting with expense management. By leveraging AI agents that can autonomously reason, act, and enforce policies, Ramp significantly reduces administrative overhead. The company learned that rather than building many single-function agents, focusing on a unified agent with diverse skills creates a simpler, more scalable system. Central to success has been an iterative approach grounded in careful data labeling, continuous evaluation, and building AI infrastructure that supports both developer and organizational agility. Beyond technical innovation, Ramp highlights the profound cultural adjustments necessary across product, engineering, and finance teams to deliver impactful AI features rapidly and responsibly.

This content is essential for product managers, engineers, and AI practitioners building AI-powered automation tools, particularly in finance and operational workflows.