JUG.CY, Jun 12, Build agents with Koog and debug by agents with AppGlass on JVM

Join JUG.CY for “Build agents with Koog and debug by agents with AppGlass on JVM”, a meetup focused on two technologies that make AI on the JVM simpler, more practical, and more productive. Expect two deep technical talks with real Java and Kotlin code: don’t let the word “agent” mislead you — this won’t be vibe coding, it will be bytecode, JVM internals, and hands-on engineering.

Participation is completely free, but places are limited. Please check in if you plan to attend the meeting; we can't let you into the office if you are not on the list.

Agenda for this evening:

- Doors open at 18:30
- Talks will start at 19:00
- We will have a short break after the first talk
- We will have discussions regarding the talks with networking opportunities after.
- We will finish before 22:00 at the office, but nothing can stop us from continuing somewhere in the cozy bar.

Talks:

Vadim Briliantov - Why Most AI Agents Never Scale? Building Enterprise-Ready AI with Koog

I agents are everywhere — but most of them break the moment you try to use them for anything real. Token costs explode, behavior becomes unpredictable, and primitive "LLM + loop" architectures simply don't scale beyond flashy demos.

At JetBrains, we've been building agents that power real products used by millions. Koog is the open-source Kotlin and JVM framework that came out of this experience and was released at the KotlinConf 2025 exactly 1 year ago. In this talk, we'll introduce Koog 1.0.0 and show how its design makes AI agents scalable, predictable, and ready for production.

Koog gives Kotlin and Java teams a complete toolset for building agents as well-structured, type-safe systems — whether you're using simple functional agents, graph strategies, or planning-based approaches. It integrates deeply with the JVM and Kotlin Multiplatform ecosystems, including Spring, Spring AI, Ktor, Langfuse, W&B Weave, AWS Agent Core, Google Agent Engine, and full support for Android and Gemma-based local agents.
We'll also look at how Koog handles challenges that most agent frameworks avoid:
Managing cost and context at scale through strategy-driven decomposition.
Modeling domain behavior using strongly-typed steps, so agents produce reliable, controlled outputs instead of guesswork.
Persisting and checkpointing agent state for fault recovery and long-running workflows.
Observing, evaluating, and improving agents with OpenTelemetry, Langfuse, and W&B Weave.

By the end, you'll understand why most AI agents don't scale and how Koog helps you build the ones that do: agents that run across JVM and KMP targets, integrate cleanly with existing systems, and remain robust under real-world load.
If you want to bring AI agents into production without rewriting your stack or sacrificing reliability, this talk is for you.

Nikita Koval - AppGlass: Non-Suspending and Safe Tracepoints for Production JVM Applications

Traditional production debugging forces a trade-off: either collect indirect signals like logs and traces, or attach a powerful debugger that can interfere with the running system. AppGlass — a new product from the JetBrains startup incubator — explores a different approach: JVM tracepoints with runtime snapshots and user-defined conditions designed to be safe enough for production use.

In this talk, we’ll show how AppGlass works under the hood: JVM bytecode instrumentation, snapshot capture, conditional tracepoint evaluation, and streaming tracepoint hits back into IntelliJ IDEA. We’ll focus on the engineering challenges behind production-grade live debugging — especially how to execute developer-defined conditions while guaranteeing they cannot mutate application state or otherwise affect the observed program. We’ll also show how AppGlass can be used both with AI agents and for debugging local code and tests, giving developers and agents access to real runtime data.

About the speakers:

Vadim Briliantov (@vadim.briliantov) Technical Lead and creator of the Koog framework at JetBrains.
Over the past 8 years at JetBrains, Vadim has contributed to a wide range of projects, including the IntelliJ Kotlin plugin, Kotlin Libraries, the Ktor framework, the Qodana Cloud Backend, and Kotlin Multiplatform Tooling. He has also led the technological direction of AI agent development across multiple products as part of the AI Agents Platform. Currently, he leads the development of the Koog framework.

Nikita Koval (https://nikitakoval.org/), Technical Lead for the Debugging Research team at JetBrains and specialize in concurrent programming and program analysis. Recent highlights include redesigning the synchronization and communication primitives for Kotlin coroutines and creating Lincheck, a testing framework for concurrent data structures. His primary research interests include but are not limited to concurrent data structures and algorithms, their verification, and practically applicable code analysis.