Orange Business Builds Live Intelligence Platform on LangChain and LangSmith

Orange Business, the B2B arm of Orange Group, built its flagship AI platform Live Intelligence using LangChain and LangSmith to orchestrate complex enterprise workflows. The platform is currently deployed across 120,000 Orange employees and has been adopted by 150+ customers in France and Europe.

Architecture and Components

The platform consists of three distinct components targeting different user roles:

• Live Intelligence Studio: An environment for AI engineers to design cross-system workflows, featuring an LLM-as-a-Service catalog, an agent builder powered by LangSmith Fleet, and MCP support.
• Live Intelligence Portal: A interface for business end users (HR, legal, software teams) to access deployed workflows for tasks like drafting job descriptions or generating test scripts.
• Live Intelligence Console: A control layer for enterprise administrators to monitor usage, track performance, and enforce security policies.

Governance and Evaluation

To manage scale and security, Orange Business implemented Agent and MCP Registries. These allow teams to discover, reuse, and version agents and prompts centrally, while providing a controlled catalog of trusted tools and connectors for accessing enterprise systems like Microsoft 365 and CRMs.

The team utilizes LangSmith for a three-layer evaluation stack in production:

1. RAG quality: Assessing retrieval recall, grounding quality, and answer faithfulness.
2. Agent behavior: Monitoring tool selection accuracy, task completion, and robustness against prompt injection.
3. Production trace monitoring: Tracking latency, cost, model usage, and failed runs.

Operational Impact

By leveraging LangSmith’s observability capabilities, Orange Business avoided an estimated 12 to 18 months of internal development time. The shared visibility into execution traces allows engineers, product managers, and QA to collaborate on debugging and regression testing, enabling continuous improvement of agent behavior before deployment.