An LLM on its own generates only text. What does it take to transform it into an agent capable of acting, self-correcting, and completing complex tasks? The answer lies in the harness: the external loop that connects the LLM with the environment, tools, and the real state of the system. In this talk, we will explore why an agent's true reliability depends not just on the model's capabilities, but on how its runtime is designed. Then, we'll dive into the code and build our own agent harness together from scratch.

As Artificial Intelligence becomes increasingly widespread, organizations need to determine how to govern its adoption, allocate roles and responsibilities, and balance control with innovation. This session explores the key challenges at the heart of AI Governance today: from defining organizational structures and coordinating across functions to overseeing AI usage, managing risks, and building a governance model capable of supporting AI adoption at scale.

Insights from the research of the Startup & AI Thinking Observatory, followed by a roundtable with leading Innovation Managers

Most AI software is still limited to copying and pasting responses. However, a symbiotic integration is possible, capable of processing data far more complex than PDFs or Excel sheets. The EngIQ case proves it: the platform queries 3D BIM models using natural language to extract design and strategic insights, natively integrating into operational workflows with a human-in-the-loop approach. A session designed for those who want to bring AI into complex software, beyond the simple chatbot.

Using AI through cloud APIs means dealing with usage-based costs that are difficult to predict and company data leaving the organization. With today's hardware (a Mac Studio, an NVIDIA DGX Spark, or a laptop with a recent GPU), it is possible to run private AI models on-premises: the real obstacle is complexity. Gabriele Fronzé demonstrates how Electros LaunchPad installs models and services such as n8n and Open WebUI with a single click, and explains why switching from usage-based spending to a predictable investment makes financial sense for companies.

LLMs are masters of correlation, but decisions require causation. What truly happens if I raise a price, launch a promotion, or switch suppliers? Causal AI promises to equip agents with the ability to estimate the impact of their actions, measure uncertainty, and learn from outcomes. A journey into the next enterprise AI leap: from the agent that executes to the agent that can justify its choices, moving from rule-based governance to evidence-based governance.

Artificial intelligence has made information, analysis, strategies, programming, and content accessible to everyone. If anyone can get good answers in a matter of seconds, what still makes a human being valuable?The answer lies in experience, judgment, and the ability to provide direction. Because in complex work, one word keeps coming up: “it depends.” It depends on the context, the people, expectations, resources, and timing. AI can suggest many plausible solutions. Experience helps us understand which one truly makes sense in a given situation. And human value also emerges in the ability to guide someone through times when results are slow to arrive, doubts arise, and cold strategy meets reality.Branding is changing, too: what will matter less and less is what we know, and what will matter more and more is the perspective we have developed, the trust we inspire, and the way we interpret complexity and nurture relationships.

In a context where AI makes legal knowledge accessible and replicable, the value of a lawyer shifts from acquired knowledge to the ability for continuous learning. This session explores the transition from a "know-it-all" to a "learn-it-all" culture, analyzing how to redefine professional identity, client relationships, and the very sense of expertise. It will also delve into the impact of this paradigm shift on the debate between generalist and hyper-specialized lawyers.

In modern agentic architectures, much of the planning, coordination, and delegation decision-making is delegated to language models, often introducing unnecessary computational costs, latency, complexity, and, above all, slowness. But what if the true evolution of multi-agent systems lies not in making every agent smarter, but in making the way they are orchestrated smarter? Drawing on the principles of System 1 and System 2 cognitive architectures, we will explore a methodology for designing adaptive orchestrators capable of combining real-time decision-making processes—driven by lightweight models and specialized algorithms—with deep-reasoning mechanisms activated only when needed. Through architectural patterns, implementation examples, and a comparison of different orchestration strategies, we will analyze how to dynamically manage task routing, computational resource allocation, and escalation to specialized agents or human supervision. The goal is to demonstrate how to transition from simple systems that link AI agents to architectures capable of orchestrating diverse forms of intelligence, optimizing performance, costs, and reliability.

From diagnostics to fertility, from veterinary health to cosmetics, four female-led startups show how artificial intelligence can address concrete needs, personalize products and services, and unlock new market opportunities, turning scientific and technological innovation into well-being.

How much return does AI actually deliver in a business? We answer with real-world case studies, not slides about the future: companies that implemented it, what worked, where they made mistakes, and what they learned. Using a two-axis framework—AI Adoption vs. Business Impact—we highlight four company profiles along with our recommended approach for each, from first steps to measurable results. Visit the AI in Enterprise booth to find out which profile your company fits into in just 3 minutes.

As AI evolves from generating information to executing autonomous actions, AI Harnesses emerge as the critical layer for orchestration, governance, and human oversight. Drawing on Systems Thinking, the keynote explores how AI Harnesses and Tech Diplomacy strengthen Tech Sovereignty, strategic autonomy, and accountability. The central argument is that Decision Authority is the ultimate leverage point determining who controls the rules governing intelligent systems.

How to build collaborations, leverage expertise, and share resources to accelerate technology transfer, transforming scientific knowledge into competitive products, services, and new businesses. At the center, the role of relationships and synergies among ecosystem stakeholders in bridging the gap between lab and market, and in strengthening the capacity of territories to generate AI-driven innovation, growth, and international competitiveness.

AI arrives in our workplaces carrying extraordinary promises: new opportunities, greater productivity, and entirely new ways of working. But it also brings new responsibilities and a powerful wave of human emotions:  Excitement from managers envisioning unprecedented opportunities.; Frustration from IT teams facing skyrocketing expectations after non-technical colleagues vibe-code a prototype in an afternoon: “If I built this in a few hours, why does IT need months?”; Fear from employees wondering whether their skills and jobs will still matter tomorrow; Pride and relief from people discovering how AI can make their work easier, sometimes mixed with anxiety about their own future.  Sounds familiar? So, what can we do about it? These emotions interact, spread across teams, and shape the way organisations embrace AI. Understanding their hidden patterns can reveal valuable signals about our expectations, relationships, and the future of work. In this session, we will explore the emotional architecture of AI transformation and offer a practical gameplan for AI adoption, grounded in research and current best practice, with actionable lessons on how to turn empathy into organisational capability.

As autonomous agents dominate headlines, the broader landscape of artificial intelligence is often flattened into a single paradigm: throwing frontier LLMs at every problem. Yet truly orchestrating intelligence requires knowing the full territory, from optimization, graph reasoning, and classical ML to foundation models and multi-agent architectures. This session charts the technical spectrum of AI methodologies against two critical engineering axes: real-world use-case geometry and vital safety constraints. We will explore how different AI paradigms fail, where they complement each other, and how to combine deterministic, statistical, and generative techniques into hybrid systems. Attendees will walk away with a clear architectural decision framework for selecting, combining, and governing the right forms of intelligence for the right problems.

Artificial intelligence steps out of screens and takes shape in robots, intelligent machines, and autonomous systems capable of perceiving, interacting, and acting in the real world. How to orchestrate AI, mechatronics, and industrial expertise to transform research into concrete applications, accelerate technology transfer, and position Italian innovation as a leading player in the new industrial revolution?

AI is transforming what we can do. The next leadership revolution will be defined by what we dare to decide — and what we choose to stand for. We are entering a defining era of artificial intelligence. Autonomous agents, intelligent systems and human expertise are converging, fundamentally reshaping how organizations operate, how decisions are made and how value is created. But beyond the extraordinary capabilities of AI lies a more profound leadership challenge: how do we orchestrate increasingly powerful forms of intelligence while preserving the human qualities that give our decisions meaning? The next AI leadership frontier is not simply about technological acceleration. It is about redefining leadership itself. Drawing on more than 25 years of international experience in technology, transformation and multicultural leadership, Ariadna Grobério explores the emergence of the AI-augmented leader and the changing relationship between intelligence, judgment and responsibility. Inspired by the art of conducting a symphony, this keynote challenges conventional thinking about leadership in an increasingly autonomous world. It examines why empathy, courage, ethical judgment and accountability are not secondary to technological progress, but fundamental to its success. Moving beyond automation and efficiency, the session offers a compelling vision of a future in which human and artificial intelligence work in harmony — guided by purpose, grounded in shared humanity and directed towards meaningful progress. Because technology can amplify intelligence. But only leadership can give it direction, responsibility and purpose.

The competitive potential of Italian AI/SaaS startups: technological quality, entrepreneurial capability, scalable business models, and the role of investors and ecosystems in turning innovation into international leadership.

We have learned to ask AI for better answers. But what happens when AI begins to influence even the questions we are capable of asking? LLMs, search engines, and recommendation systems are becoming a true cognitive infrastructure: they select information, introduce concepts, suggest connections, and help construct the space of possibilities in which our thinking is born. The next leap in skills will therefore not be prompt engineering alone. We will be asked to learn how to preserve our cognitive autonomy: recognizing assumptions, seeking what is missing, comparing alternatives, and formulating questions that the system has not suggested to us. If in cybersecurity we perform hardening of digital infrastructures, in the AI era we will also need to learn how to perform hardening of thought. When answers are infinite, the human value lies in still being able to imagine a different question. And open up new possibilities.

Most failures in production AI agents are not purely model failures. They are often context failures: stale retrieval, bloated prompts, inconsistent memory, recursive tool loops, and hidden state spread across vector stores, chat history, workflow systems, and agent orchestrators.  This session presents a systems architecture approach to agent memory. We will break down the main memory layers in modern AI agents, including short-term context, episodic memory, semantic memory, procedural memory, and external state. Then we will examine why more context does not always improve reliability, and how unmanaged context can increase latency, cost, and failure rates.  The core of the session is a practical framework for controlled forgetting. Drawing from distributed systems patterns such as cache eviction, TTLs, log compaction, snapshotting, and state reconciliation, attendees will learn how to design agents that remember the right information, discard stale or irrelevant state, and remain observable as workflows scale.  Attendees will leave with a reference architecture for context engineering, concrete failure modes to monitor, and implementation patterns for building reliable, cost-aware, stateful AI systems.

An exploration of how to orchestrate innovation, expertise, and investments to build a smarter, more transparent, and resilient food supply chain.

AI’s biggest promise isn’t doing yesterday’s work faster. It’s making new work possible. Drawing on his work across enterprise AI and creator partnerships, Edward Frank Morris shows how different forms of intelligence can be brought together to make that happen. From teaching AI to government teams in Ukraine and working with OpenAI, Perplexity, Ant Group, Alibaba and Tencent, to creating with tools from Runway, ElevenLabs and Fish Audio, he has seen AI play very different roles: colleague, researcher, production team and sometimes a rather enthusiastic intern.  The lesson is that the model is only one part of the system. The real work is deciding what matters, giving each capability the right job, and keeping human judgement in the loop. Because orchestration isn’t about having more AI. It’s about making something possible that wasn’t possible yesterday.

In many organizations, AI is still approached primarily as a tool to optimize existing processes and consequently cost cutting. But what if we are simply making the present more efficient, without adapting our organizations to the future? without understanding that this is a revolution and not an evolution. This talk challenges a fundamental assumption: the real transformation driven by AI is not technological — it is organizational, cultural, and deeply human. As AI reshapes how work gets done, we are witnessing a shift from static roles to fluid capabilities, from predefined tasks to dynamic problem-solving, and from individual contribution to human–AI collaboration. At the same time, the nature of skills is evolving faster than ever, making lifelong learning not a choice, but a core organizational capability. Yet most companies are still structured around models of work, leadership, and learning designed for a different era. This session offers a forward-looking perspective on how work is evolving globally, connecting three critical dimensions: Leadership, as the ability to navigate uncertainty and enable new ways of working Organizational culture, as the real enabler — or blocker — of AI transformation Learning, shifting from episodic training to continuous, embedded development So if we do not change our society AI is going to change us and we will not even understand it. The key question is not how to implement AI in today’s organizations, but whether we are ready to redesign organizations for the age of AI. This talk is for leaders, HR, and decision-makers who want to move beyond efficiency and start rethinking work itself.

Querying a model is simple; orchestrating the interaction between an operator and a machine opens up an entirely different perspective. Conversational analytics becomes the control center for operational processes, bridging human intelligence and action: it deciphers the big picture through the evolution of dialogue, triggers enterprise systems, enforces AI Act safeguards, and places the final decision back into human hands. From tool orchestration to seamless, secure collaboration.

We are already immersed in a world where artificial intelligence agents can select, negotiate, and purchase on our behalf. But what does buying really mean for a human being? Even the most mundane purchase can be an act through which we assert our identity, exercise our autonomy, and impact the world around us. I buy, I provide, I resolve. Reflecting on agentic commerce therefore means exploring something that goes far beyond retail: the boundaries of collaboration between humans and artificial intelligence. Where does the technological limit end, and where does the boundary tied to our identity begin? Will we truly be ready, in the name of convenience, to hand over our identity-defining functions to technology? Agentic Commerce thus becomes a testing ground for understanding not only the sustainability of the AI business model and the future of shopping and marketing, but also how our overall relationship between humans and technology will evolve.