Cover Image for GenAI Israel - AI Agents in Production: Skills & runtime execution
Cover Image for GenAI Israel - AI Agents in Production: Skills & runtime execution
Avatar for GenAI Israel
Presented by
GenAI Israel

GenAI Israel - AI Agents in Production: Skills & runtime execution

Registration
Past Event
Welcome! To join the event, please register below.
About Event

Generative AI builders, engineers, and tech leaders!
Join us for the final GenAI Israel community event of 2025 — an evening of cutting-edge talks, practical insights, and community networking at Vega's office.

🎙 Agenda Highlights:

Skills & code driven AI Agents: Building effective AI-native SecOps

Deploying AI agents in production requires a shift from monolithic prompt engineering to robust system design. This talk details how Vega developed an AI-native SecOps platform leveraging content-as-code, modular agent skills, Small Language Models (SLMs), and secure code-execution environments to achieve high accuracy with low latency. 

We will cover:

•⁠ ⁠Skills-Driven Architecture: Transitioning from long-form prompts to modular task execution.

•⁠ ⁠Declarative Frameworks: Using declaratively defined content to ensure agents can create, edit, integrate and deploy valid content effectively.

•⁠ ⁠Security Log Analysis: How we leverage SLMs for efficient, high-speed processing in large scale environments

About Gili
Gili leads AI strategy and execution at Vega. He previously spent nearly a decade in the IDF (Unit 8200) building production-oriented ML/AI products and data pipelines.He holds an M.Sc. in Computer Science from the elite Havatzalot program.

ATP: codemode runtime execution

AI agents reason well but fail on real workloads. The limiting factor is execution, not model capability.

This talk presents a runtime-first architecture for AI agents. We show why prompt-based tool calling fails when agents must process large datasets, perform multi-step computation, or interact with complex systems, leading to accuracy degradation, latency issues, and high cost.

We describe how sandboxed runtime execution with fine-grained security policies enables agents to fetch, compute, and act on real data under explicit control, making them viable components of production systems.

The talk provides a concrete mental model for designing AI agents that execute logic instead of hallucinating.


About Gad & Gal
​•⁠ ⁠Gad Shalev, Tech Lead - CEO Office Member @ Monday.com | ATP Co Creator
​•⁠ ⁠Gal Liberman, Senior Tech Lead @ Monday.com | ATP Co Creator

Location
vega.io
Beit Rubinstein, Lincoln St 20, Floor 25, Tel Aviv-Yafo, Israel
, חניה -החניה הקרובה ביותר היא בסנטרל פארק המלאכה (המלאכה 4) ת׳׳א ניתן להגיע ברכבת קלה, קו אדום תחנת קרליבך
Avatar for GenAI Israel
Presented by
GenAI Israel