LangGraph Helps AppFolio Double Agent Accuracy and Save 10 Hours a Week

Description and use

LangGraph doubled AppFolio’s agent decision accuracy and saved over 10 hours per week, while Uber automated test generation with stateful resilience.

Problem

AppFolio needed more accurate agent decisions and reduced latency for property managers, while Uber required durable multi-step test generation without state loss.

Outcome

AppFolio doubled agent decision accuracy and saved over 10 hours per week; Uber automated unit test generation with stateful resilience.

Workflow

• 1. Model agent execution as a directed graph with nodes and edges.
• 2. Add deterministic steps (validation, data retrieval) and LLM-driven nodes in the same graph.
• 3. Enable checkpointing after each node to preserve state.
• 4. Integrate LangSmith for tracing, evaluation, and real-time debugging.
• 5. Deploy with a distributed backend (e.g., ScyllaDB) for multi-region durability.
• 6. Use LangGraph Studio for visual debugging and time-travel inspection.

Evidence

AppFolio doubled agent decision accuracy and saved property managers more than 10 hours a week.

Source

Uber automated unit test generation without losing context mid-run.

Source