// Disrupt (AI) Summit Series · Agenda
Two days of inspiring keynotes, hands-on workshops, and new connections at the premier summit for Artificial Intelligence in Greece. Switch between the Conference track and the instructor-led Bootcamp for each day below.
George Nikolaropoulos sets the stage for the next two days, highlighting how Generative AI is transforming product development, engineering, and leadership. We spotlight the most pressing opportunities and challenges in AI today — from unlocking user insights and predictive analytics to automation, personalization, and responsible AI — your first glimpse into the conversations, workshops, and case studies that shape the event.
Most AI initiatives fail not from lack of innovation, but from building solutions before defining success. This session introduces the Reverse Strategy Framework — an evaluation-first methodology that flips traditional AI development on its head. Instead of selecting tools first, attendees learn to define measurable outcomes, establish evaluation criteria, and architect autonomous systems that deliver business value at scale. Drawing on enterprise deployments, we dissect why demos rarely reach production and reveal the hidden data, decision, and evaluation debts that sabotage implementations — leaving you with a playbook for multi-agent orchestration, production observability, and ROI measurement.
This session explores how the integration of traditional machine learning with AI agents is transforming core market research — enabling faster insight generation, more advanced audience segmentation, and precise, real-time quota optimization. It shows how these advances boost both the efficiency and reliability of research, redefining how organizations operate and make data-driven decisions at scale.
AI is writing more code than ever — but most of it never reaches production. The real challenge isn't capability; it's trust. Modern applications are inherently distributed, creating edge cases that AI struggles to anticipate and traditional testing fails to uncover, especially when system contracts stay implicit rather than verifiable. Through a real-world distributed app built with Gen AI, we examine the patterns, tools, and infrastructure that let AI-assisted development scale safely — and look ahead at frameworks that make whole classes of distributed failures impossible to express, pointing toward systems that are correct by design.
The PM role is undergoing its most significant transformation since Agile. As AI agents absorb execution, product leadership shifts from managing backlogs and writing tickets to designing systems, engineering intent, and making the judgment calls no model can make. Drawing on real experience building AI-native products and directing agent systems in production, this talk delivers six hard-won lessons — no theory, just patterns visible only from the other side of the transition — plus a clear mental model of the Orchestrator PM and concrete steps to begin the shift before it's made for you.
Every AI agent you ship has enemies — yours just hasn't met them yet. Quarterly pentests, bug bounties, and waiting for a customer complaint don't survive a system that ships weekly. In the agentic era the only security discipline that scales is one you run yourself, continuously, inside your own development cycle: you build the adversary, run the attack in a controlled environment, and your agent learns what it can't handle before it meets a real threat. Every break becomes a rule; every rule becomes runtime defense; human judgment governs what ships.
Participants explore how to unlock the full potential of Claude for building enterprise-grade AI applications. We cover practical strategies for integrating Claude into real-world workflows, extending its functionality with custom tools, and implementing Model Context Protocol (MCP) servers to enable scalable, secure, production-ready AI systems. Designed for technical leaders, product teams, and AI practitioners deploying Claude-powered solutions at scale.
Last year, LLMs helped developers write code faster. This year, Agentic AI and the Model Context Protocol (MCP) are transforming how applications are built end to end. See how Agentic AI accelerates the path from prototype to production by collaborating across design, frontend, backend, and cloud workflows — through a live walkthrough showing how developers use agents and MCP to build, test, and deploy production-ready applications.
The next wave of ecommerce won't be driven by better interfaces — it will be driven by AI agents that shop, compare, negotiate, and pay on behalf of consumers and businesses. This shift from “click-to-buy” to “intent-to-done” will rewire how merchants sell, how payments flow, and how trust is established in digital commerce. Maria explores how agentic commerce moves beyond chatbots into a world where AI agents become the primary counterparty in transactions — and what that means for merchants, payment providers, and the infrastructure connecting them.
Wrap up Day 1 with key takeaways from the development-focused sessions, hands-on coding, and deployment labs. Celebrate standout projects, reinforce learning, and get ready for Day 2's advanced workshops and deep dives.
The morning sessions lay the foundation for designing AI-powered products. Participants align on teams, roles, and project prompts, then dive into the fundamentals of AI product management and problem discovery. After a unified kickoff and orientation, attendees explore AI PM fundamentals — reasoning, pattern recognition, summarization, generation — define users and pain points, and identify where AI delivers real value while surfacing technical, data, behavioral, and ethical risks. They close by designing AI-driven experiences: user-centric design, human-in-the-loop flows, fallback states, and scoping a feasible, valuable MVP. By midday, teams are aligned with a clear problem statement and a first blueprint for AI-driven product ideas.
This block helps participants translate design thinking into actionable AI product concepts, combining user experience, metrics, risk management, and MVP planning. Participants transform problem framing into a concrete user experience — mapping the journey, defining AI touchpoints, planning for ambiguity and failure modes, designing both the happy path and critical edge cases. They learn to measure success by aligning user outcomes with business impact, distinguishing product vs. model metrics, and managing data and operational risks. Finally they consolidate everything into a structured, pitch-ready AI product concept. By the end, participants have a fully defined, realistic, and ambitious concept ready for prototyping and presentation.
Despina Kitsou opens Day 2, framing the developer-focused track and the deep technical workshops ahead — setting expectations for the hands-on coding, agentic patterns, and production-grade AI engineering that define the second day of Gen AI Unfold.
Legacy systems aren't the real obstacle — outdated thinking is. This session showcases how Natech leveraged AI to transform a traditional SQL database project into an AI-enhanced codebase, unlocking faster development cycles, stricter quality controls, and smarter, automated bug handling — all without rebuilding from scratch.
Many organizations have launched GenAI pilots; far fewer have built the architectural and governance foundation to scale them into secure, reliable, business-aligned capabilities. As enterprises move from assistants toward agentic systems that reason, orchestrate, and act, the challenge becomes platform design, integration, observability, and control. This session presents a practical blueprint for the full journey — identifying the right use cases, defining architecture patterns, enabling secure integration, introducing governance guardrails, and operationalizing trust through observability and human oversight. Attendees leave with reusable artifacts: a reference architecture, a governance checklist, and a step-by-step scaling framework.
Integrating Generative AI into everyday workflows is becoming a core product leadership capability. This session presents a real-world transformation of a multi-thousand-person Product Management organization — moving from an org-wide AI strategy to agent-enabled workflows embedded in daily operations. Participants gain a structured, step-by-step framework for analyzing existing product-creation processes and systematically introducing GenAI across the lifecycle: discovery and user research, ideation, experimentation, roadmap planning, and go-to-market — with concrete use cases and clear starting points to raise productivity, decision quality, and learning velocity.
You may have seen headlines about AI systems recommending people eat rocks or put glue on pizza — classic hallucinations. Retrieval-Augmented Generation (RAG) grounds model outputs in reliable external knowledge to reduce these failures, combining a retriever (typically a vector database) with a generator (an LLM). The concept is deceptively simple, but production RAG introduces real-world challenges across data quality, retrieval accuracy, latency, evaluation, and operational complexity. This talk explores the most common failure modes and the practical techniques to mitigate — or eliminate — them, for more reliable, trustworthy AI applications.
By 2026, talent scarcity is no longer a hiring problem — it's a structural risk. While 85 million roles may go unfilled by 2030, the deeper issue is organizations' inability to deploy skills quickly and flexibly under pressure. This session reframes workforce optimization around team design, showing why narrowly specialized teams collapse under disruption and why multifunctionality and coverage are strategic necessities. Attendees meet an AI-driven workforce intelligence approach that models team reliability and resilience, simulates disruption scenarios, and pinpoints where targeted skill development delivers the highest return — a practical framework to redesign how work is structured before scarcity becomes operational failure.
Building an intuitive, self-updating, scalable AI legal assistant for a complex, highly regulated environment like Greece raises challenges beyond typical product development. This session examines the evolution of dikaio.ai — the first AI legal assistant for the Greek market — focusing on the practical lessons of turning an early concept into a tool used in daily professional workflows: selecting the initial user segment, adapting to law firms, financial institutions, and the public sector, and shifting from a conversational interface to a system integrated into critical operations. Special attention goes to how product–market fit was validated in practice through sustained usage and renewals.
AI voice agents look simple in a demo and collapse in production — and the reasons span every layer. The LLM breaks (hallucinated policies, upgrade regressions, accent gaps, misread intents). The telephony layer breaks (SIP quirks, failed transfers, voicemail detection). The knowledge layer breaks (stale FAQs, missing edge cases). The action layer breaks (wrong CRM field, broken workflow). And the feedback loop is usually missing entirely. This talk introduces a five-layer framework — Scope, Voice, Brain, Hands, Loop — mapping every production voice agent onto the same anatomy, with each layer's failure modes, build order, and minimum-viable version, drawn from two years running voice agents across collections, e-commerce, booking, and support.
Unlock the true potential of AI by mastering the language it understands best — prompts. In this lab session you'll learn to design clear, effective, impactful prompts that drive accurate and creative responses. We explore the principles of prompt engineering, common pitfalls to avoid, and practical frameworks to structure your requests for maximum results. Whether you're a beginner or refining your skills, you'll leave with actionable strategies to communicate with AI confidently and efficiently.
Despina Kitsou closes Gen AI Unfold 2026 — consolidating the two days of keynotes, workshops, and bootcamps into the key takeaways attendees carry forward into their own AI work.
A one-day, hands-on technical bootcamp for engineers diving into the AI era. Participants learn LLM fundamentals, set up local models (Ollama, LM Studio), and build their first generative AI web app using TypeScript, the Vercel AI SDK, Next.js, and Shadcn UI — covering Retrieval-Augmented Generation (RAG) and tool integration so attendees can code and deploy their own GenAI apps in a day, all running on their local machines. The morning begins with a unified kickoff, then an LLM fundamentals deep dive (tokens, context windows, temperature, prompting), and closes with local LLM setup and configuration — performance, privacy, and model trade-offs. By midday each participant has a fully functional local LLM ready for experimentation.
The afternoon sessions focus on building end-to-end GenAI applications by combining AI backends with modern frontend frameworks. We start with TypeScript + Vercel AI SDK fundamentals — SDK architecture, API patterns, streaming responses, and prompt engineering — culminating in a working “Hello World” GenAI app. Next, in Next.js + Shadcn UI frontend development, participants build a modern, reusable frontend connected to their AI backend, emphasizing project structure, state management, UI/UX best practices, and integration, delivering a functional frontend scaffold with chat UI ready for further development.
The final sessions focus on advanced AI workflows and real-world deployment. In RAG & Tools Integration, participants learn Retrieval-Augmented Generation fundamentals — embeddings, vector databases, retrieval workflows — while exploring tool calling and agentic patterns that make applications more context-aware and reliable. During the Build & Deploy lab, attendees integrate their local LLM, RAG pipeline, and tools into a fully functional GenAI application ready for deployment. The day concludes with a live demo showcase where selected participants present their applications, share insights, receive feedback, and celebrate achievements — fostering community and peer learning.