Nemanja Petrović, Principal @Databricks

This talk covers the current technological and economic landscape in high tech, the challenges and opportunities it presents, and the disruptive potential of AI — for businesses broadly and software companies specifically — followed by practical guidance and speculative solutions to use AI to expand the pie, not cut the costs.

Srđan Božović, Microsoft

This talk is about using LLMs, retrieval, and agents to change how large software systems are actually built, tested, and operated. Drawing on our experience applying AI across the Azure SQL engineering stack, I’ll walk through a range of real problems we’ve attacked: making LLMs reason reliably about massive codebases, grounding them with vector and graph-based retrieval, building agents that take on concrete SDLC tasks (code comprehension, test authoring, flaky-test triage, livesite investigation, code review), and measuring whether any of it actually moves the needle. Expect architectural patterns, honest failure modes, and a frank view of where current models shine, where they quietly mislead you, and where the interesting engineering work still lies. Useful whether you’re working on a startup codebase or a multi-decade one.

Dragan Pleskonjić and Vladimir Jelić, Glog.AI

This session introduces a necessary paradigm shift: moving from reactive incident response to a proactive, predictive security framework. We will demonstrate how AI-powered solutions can anticipate threats before they emerge, turning complexity into clarity. This approach will be explored through three core pillars of a unified security ecosystem:
• Glog.AI – Powered Software Security and DevSecOps: Automatically identifying and remediating vulnerabilities in real-time during the development lifecycle.
• Advanced Network Security Anomaly Detection (NetEcho): Identifyng real-time threats, attacks, and exploits within network traffic using a novel, AI-driven approach. By detecting emerging attack vectors at their earliest stages, it enables organizations to move from reactive defense to proactive, optimized security planning.
• Security Predictions – Proactive Threat Intelligence: Analyzing vast datasets and signals to anticipate emerging attack vectors and patterns, enabling preemptive defense posture adjustments.
The presentation will culminate in a live demonstration of our AI-powered software security platform, showcasing its capability to automatically detect and fix code vulnerabilities in a real-world development environment.

Danijel Dejanović, Skenify

Agentic AI systems are recently associated with many security risks, from OpenClaw issues to misuse of Claude to execute serious attacks, we have to rethink how these systems can and should be positioned to serve the defenders. In this presentation I will walk you through concepts, use cases and practical applications of AI Agents in service of protecting the organizations from bad actors and giving business owners the edge a small business never had before.

Đorđe Ivković, XG Dynamics

The advent of Artificial Intelligence (AI) requires significant changes to the way that Cyber Security operations are performed. In this talk, I will be showing how AI is revolutionizing the way that network attacks have evolved both for attackers and defenders in modern network infrastructure. Using open-source LLMs, we design and build attacking (red) and defensive (red) agents for digital twins of cyber security simulation environments. To that end, build a cybersecurity simulation environment designed to model real network defense and attack scenarios. To preserve realism, agents operate under partial observability and must make decisions with limited and incrementally revealed information. Our approach demonstrates adversarial (red) agents, which attempt to explore, exploit, or compromise network targets and blue agents, which attempt to defend and secure the network. The design of the environment allows for repeatable and measurable interactions to allow for strategy extraction at scale. Moreover, we show how we track known state information about the environment to provide richer observations. Furthermore, we discuss fine-tuning techniques to improve agent performance. The session highlights the challenges and techniques involved in adapting and running open language models in complex, partially observable, game-like environments for digital twins of real, production networks to foster cyber security.

Goran S. Milovanović, Datakolektiv

AI transformation is often framed as a technology adoption problem – selecting the right tools, building the right models, integrating the right platforms. This framing is incomplete. Organizations are ecosystems of people, processes, information flows, and decision structures, and introducing AI at scale perturbs every layer of that system simultaneously.

This talk approaches AI-driven business transformation as systemic organizational change. Drawing on over two decades of work at the intersection of cognitive science, data science, and executive leadership, I examine how AI migrates from isolated pilots to a structural capability embedded across analytics, product, and business processes – and what this migration actually demands from organizations.

Key themes include: the co-evolution of technical infrastructure and organizational culture; how communication patterns, decision rights, and team topologies must adapt when AI becomes a first-class participant in workflows; the distinction between AI as a tool and AI as an integrated capability; and the feedback loops – both productive and pathological – that emerge between human expertise and machine intelligence during transformation. I will also address the governance, skills, and leadership shifts required to convert AI investment into measurable business value, rather than into technical debt or organizational friction.

The goal is to equip leaders with a systems-level lens for diagnosing where their organization actually stands in its AI transformation, and for designing interventions that address the ecosystem, not just the stack.

Tamara Zavišić, Zavisic.co

How do you move from an AI idea to a product that is not only functional, but also compliant and responsibly governed? This presentation will walk through the development process phase by phase and highlight the key compliance and governance questions that should be addressed at each step. Framed through an agile compliance approach, it will show how legal, risk, and governance requirements can be integrated into product development early and continuously, rather than treated as a final control point. Using the EU AI Act and ISO/IEC 42001 as the main reference points, the session will help participants understand how to build AI products in a way that is structured, accountable, and aligned with real regulatory and organizational expectations. In addition to the session content, participants will receive a set of basic templates designed to support initial compliance, governance, and documentation steps in AI product development.

Atila Alaćan, 4zida.rs

Most developers start by stuffing as much context as possible into their prompts. I did the same and watched my agents get worse. In this talk, I’ll show why more context often leads to less intelligence and how a precision-first approach to context engineering leads to better reasoning, reliable outputs, and agents that actually finish what they start instead of drifting off.

Goran S. Milovanović (moderator), Dragan Pleskonjić, Tamara Zavišić, Bryan Guillaume

As AI moves from experimentation into everyday systems, understanding its real impact becomes critical. This panel brings together diverse perspectives to examine where AI stands today, what use cases are proving effective, what risks are emerging, and how to think about the future without hype or oversimplification.