Codepocalypse Now: LangChain4j vs JetBrains Koog

Abstract

Two frameworks, two presenters, one increasingly capable assistant. Baruch Sadogursky and Viktor Gamov build j-claw live in JetBrains Koog and LangChain4j Agentic, using an urgent corporate problem as the demo: getting out of mandatory AI training. Across four rounds, a chatbot gains tools through MCP, persistent memory, reusable skills that turn corporate-speak up to eleven, and a typed workflow with a critic and a refinement loop. Along the way, Gemini, Claude, and Codex take different roles, and agent traces make the workflow visible. Both frameworks can get the job done; the hard part is modeling the work, giving the agent the right context, and ensuring that a rejected draft cannot slip through to delivery.