After evaluating various managed APIs and frameworks, Ramp invested six weeks to build Laminar, a comprehensive AI agent harness designed for better cost efficiency, improved reliability, and enhanced developer workflows. This cloud-native platform powers multiple fintech applications while maintaining flexibility in model selection and deeper insights into agent behavior.
- In-house harness enables cost and model flexibility
- Asynchronous agents improve workflow efficiency
- Centralized APIs enhance observability and security
Infrastructure signal
Ramp’s decision to build Laminar internally rather than relying on external managed agent APIs reflects a strategic move to reduce dependency and improve cloud cost management by allowing rapid iteration on model choice based on performance and price. The architecture separates agent session state from the sandboxed code execution environment, increasing security while optimizing resource usage by spinning sandboxes up or down without disrupting workflows.
Laminar consolidates agent execution, sandbox management, and observability into a single platform, enabling granular tracing of agent activity and simplifying debugging. By centralizing authentication, credentials, and tool orchestration behind a Connectors Service API, Ramp creates uniform control points that enhance security posture and operational consistency across their cloud environment.
Developer impact
Building the harness in-house fosters closer collaboration between platform engineers and product teams, shortening feedback cycles and enabling customized capabilities aligned exactly with business needs. Integrated observability tools provide developers with better insights into agent performance and troubleshooting information, reducing the time spent navigating third-party tools and improving reliability.
What teams should watch
Teams responsible for cloud cost and reliability need to evaluate the trade-offs between in-house infrastructure versus managed platforms, particularly focusing on how model flexibility and observability capabilities can keep expenses predictable and aligned with performance targets. Maintaining sandbox isolation and managing credentials centrally are crucial for security teams as agent complexity grows.
Product and platform engineering groups should observe how asynchronous agent management and persistent session handling transform development workflows and deployment strategies. The shift towards modular agent harnesses—as shown by both Ramp’s Laminar and LangChain’s Deep Agents—signals an industry move to more adaptive and maintainable AI infrastructure, which will likely influence API design and platform integration choices going forward.