We’re creating a space for tech minds to connect, share ideas, and talk about the real stuff: new trends, head-scratching challenges, or those quirky edge cases and gotchas that only fellow techies understand.
Forget long, one-sided presentations. These sessions are about knowledge-sharing with like-minded people. Talking. Exchanging. Relaxing.
February talk: AI Agents Memory — State Of The Art For 2026
Over the past two years, the field of “agentic AI” has shifted from flashy demos to rigorous engineering: the key bottleneck is no longer raw model size or reasoning capabilities, but how agents remember – or forget. In this talk, we survey the state of the art (as of 2026) in memory systems for AI agents: what works, what still fails, and where the frontier lies.
We begin by defining what “memory” means in the context of an AI agent – distinguishing between ephemeral short-term context, session-spanning medium-term recall, and persistent long-term memory. We then review major recent architectures: from hierarchical storage systems such as MemoryOS (short-, mid- and long-term memory layers) , to scalable memory modules like Mem0 which demonstrate large gains in conversational coherence, latency, and token efficiency compared to naïve context-window approaches.
We’ll also explore cutting-edge developments from late 2025 and 2026: for example, hybrid frameworks such as PISA – a psych-inspired unified memory system for lifelong agent learning – which bring together adaptive schema evolution, hybrid neural-symbolic retrieval, and robust long-term retention. Similarly, the new multimodal memory framework MemVerse demonstrates how agents can persist and reason over multimodal (text, image, sensory) experiences – a major step toward lifelong, embodied AI.
Beyond architecture, we reflect on system-level lessons and open challenges: how to decide what should be remembered (vs. forgotten), how to prevent “memory bloat,” and how to preserve trust and safety when agents recall user preferences or personal data across sessions. Finally, we lay out a forward-looking roadmap for memory in agentic AI – including modular, portable memory layers, cross-agent memory sharing, long-horizon planning grounded in experience, and hybrid memory models combining symbolic and learned components.
Attendees will come away with a comprehensive understanding of the current “memory stack” for AI agents, a sense of which technologies are production-ready today, and which innovations are likely to shape the next generation of agents.
Eugene Steinberg: CTO, deep thinker, and passionate about building things that actually work.
Nikita Ivanov: VP of Technology/AI, product-driven innovator with over 25 years of global leadership across AI, software engineering, and digital transformation.
Special Guest → stay tuned for a surprise appearance from another voice in the AI world.
The format
A cozy gathering for about 30 people; enough to spark great conversations but still feel personal.
The atmosphere
Our new office is the perfect setting: warm, relaxed, and full of good energy. There’s always great coffee, good beer, and our amazing office manager, who makes sure nobody ever leaves hungry.
Why Dynamic Talks?
Because sometimes, the best ideas happen not in meetings, — but in the moments between.
Agenda
Arrivals, coffee, and a warm welcome
Opening words & a quick intro
Keynote talk: AI Agents Memory — State Of The Art For 2026
Quick Q&A: first reactions, fresh thoughts
Panel chat: different voices, shared ideas
Open floor: a lively, moderated Q&A
Wrap-up: a few takeaways & smiles
Happy networking time 🙂 beer, snacks & good talks