Talking Kotlin Building AI Agents With Koog
14 snips
Nov 17, 2025 Vadim Briliantov, Technical Lead for Koog at JetBrains, shares insights on building AI agents in Kotlin. He defines AI agents as dynamic problem solvers that interact with environments using Koog, an open-source framework. The discussion highlights Koog's applications, including backend automations and on-device agents, while exploring features like multi-phase coding strategies and custom tool management. Vadim also delves into history compression techniques, type safety, and the roadmap for future development, emphasizing Koog's enterprise readiness and exciting open-source potential.
AI Snips
Chapters
Transcript
Episode notes
Vadim's Path From Intern To Koog Lead
- Vadim described his eight-year JetBrains journey from intern to Tech Lead of the AI agents platform.
- He explained how internal agent work evolved into open-sourcing Koog for the wider Kotlin community.
Supports MCP And Plans A2A For Interop
- Koog supports MCP for external tools and plans A2A for richer agent-to-agent interactions, plus maintains the Kotlin MCP SDK.
- Agents can be exposed as tools or connected via MCP/A2A for cross-system integration.
Five Lines To Run An Agent
- The simplest Koog entry point requires adding a dependency, creating an AI agent with prompts, LLM connection, and tools, then calling agent.run.
- Basic usage can be achieved in about five lines of Kotlin for rapid experimentation.