Four Agent Frameworks in Sixteen Months (engineering.vasco.app)

Vasco has undergone a significant transformation in its AI agent, Gama, migrating through four different frameworks in just sixteen months to enhance its functionality in revenue analysis. Initially, Gama struggled with unreliable outputs and the cumbersome requirement of babysitting large language models (LLMs) for data accuracy. To overcome these challenges, Vasco transitioned to LangChain, then LangGraph, and ultimately the Google Agent Development Kit (ADK). Each move aimed to streamline operations, but faced setbacks, particularly with deployment inconsistencies and a steep learning curve. Most recently, Vasco adopted the Claude Agent SDK, which remarkably outperformed previous implementations by leveraging a sophisticated coding runtime. This migration not only simplified their architecture but also dramatically increased Gama's reasoning capabilities and efficiency, allowing it to produce high-quality reports without the extensive scaffolding that previously constrained its performance. This strategic pivot underscores a broader trend in the AI/ML community—adapting to quickly evolving technologies and focusing on integration rather than attachment to existing frameworks, all while acknowledging the risks and rewards associated with rapid change.