Workday cuts AI agent prototyping from months to less than a day
Workday has built an internal platform for developing, deploying and operating AI agents on its own infrastructure, using LangGraph for orchestration and LangSmith for tracing, observability and evaluations. The company says the platform reduced the time needed to produce a prototype from 4 to 5 months to less than 1 day.
- Workday built an internal platform for production AI agents.
- The platform uses LangGraph for orchestration and LangSmith for tracing, observability and evaluations.
- Prototype development fell from 4 to 5 months to under 1 day.
- The Agent Platform team grew from 4 people to roughly 30.
- Workday’s platform includes governance through its Agent System of Record.
A platform designed for enterprise constraints
Workday’s leadership set a goal in 2025 to have agents in production by the end of 2026. According to LangChain’s case study, product teams needed a system that could support rapid prototyping as well as production deployments, while meeting Workday’s security, audit, compliance and data retention requirements.
The project began with a team of 4 and covered a broad set of responsibilities. These included building a durable agent server, enforcing data access controls tied to Workday’s audit policies, creating a collaborative evaluation platform and developing an SDK that abstracted the underlying infrastructure.
Workday’s Agent Platform team serves both internal engineering teams building agents and the end customers who use those agents. The platform handles authentication, security, compliance and infrastructure so developers can concentrate on an agent’s functional requirements.
LangGraph provides flexibility while LangSmith supports evaluation
The team selected LangGraph for agent orchestration because its graph-based model allowed developers to choose how their agents were designed. LangChain says this ranged from simple ReAct loops to complex deterministic workflows, without forcing teams to use a predefined agent structure.
Workday also built an Agent System of Record, a governance layer that controls which APIs agents can access. Its platform includes a Python SDK that Alex Gatto, a Senior Machine Learning Engineer at Workday, describes as “Workday LangChain.” The SDK wraps LangChain primitives and routes them through Workday’s internal services, including an LLM gateway, a RAG retrieval service and other internal tools.
LangSmith supplies tracing, observability and evaluations, while LangChain’s agent server implementations provide support for the A2A and MCP protocols, according to the case study. Workday said this meant those protocols were already supported in production for every agent server operated by its Agent Platform team.
Prototype development falls below one day
Before the platform was available, building a prototype agent took 4 to 5 months, with no guarantee that the result could reach production. Workday says teams can now run a prototype in a staging environment in under a day, with authentication and security requirements already built in.
The platform team has grown from 4 people to roughly 30, with continued investment planned. Workday said its Python SDKs and LangChain-backed primitives also connect internal services that had previously operated in isolation, including LLM gateways, RAG retrievers, recommendation engines and prompt services.
For Workday’s developers, the result is a unified interface for building agents across the company’s internal infrastructure. Alex Gatto called the partnership with LangChain “critical to our agent development” and said Workday could not have delivered value to its customers without LangSmith’s “speed and maturity.”
