We have all seen them — the hastily retrofitted chatbots and assistants that have metastasised across almost every app. And by now we know where they are genuinely helpful and where they are a nuisance. Their most striking problem is that their all-or-nothing approach turns copilots into autopilots: after the initial prompt, the user loses control. Too often the result is slop, and in many contexts, it encourages deceiving potential customers or readers. But a different generation of tools is emerging. Rather than a copilot, we should think of AI as an exoskeleton for our mind: something that makes us more capable and more precise, while keeping us firmly in control. At Thesify, we are building AI tools for academia, a domain with high standards of authorship, accountability and transparency. That constraint has forced us to design AI around the ways our users actually work, not the other way around. In this talk, I will illustrate these points using examples of first-generation AI tools that work and ones that don't. I will then show simple principles that underlie the next generation of AI tools that are quickly gaining popularity. Some of these examples are from our own customer research at Thesify and some are from other tools. I will conclude with an outlook over the main trends that we can see in AI tools.

Enterprises are piloting AI chatbots, but most projects stall at FAQ and RAG. The assistants answer questions, yet cannot reliably execute work across real services. They often lack memory, proactivity, and are stuck in silos. In this talk, I present an agentic execution layer for production AI agents. We show how to orchestrate a multi-agent system that turns your APIs and workflows into domain agents, adds a persistent memory layer, and provides observability, replay, and regression testing so teams can ship changes with control. Key takeaways: A reference architecture blueprint for enterprises. A platform for observability, tracing and versioning.

The future of decision-making will be won by organizations that deliberately pair human judgment with AI at scale: decision flows where machines grind through the data and people bring expertise, nuance, and accountability.

This session is for executives and senior leaders who don’t need to code models but do need to own the impact of AI‑infused decisions. We will unpack how data becomes wisdom through four layers, giving participants a clear, practical mental model for designing robust Human‑AI decision systems that are reliable, auditable, and safe to use in the boardroom.

As organizations increasingly rely on AI systems to support collective decisions, a key question emerges: how do AI agents resolve conflicts when there is no “right” answer? In this talk, I share insights from nearly 500 automated negotiation experiments involving leading AI models (Claude, GPT, and Gemini). The agents faced real game-theoretic dilemmas with conflicting preferences, voting paradoxes, time pressure, and institutional consequences for failure. Rather than focusing on who “wins,” the experiments reveal something more important: most successful negotiations were driven by agreement on process, not on outcomes. Over 85% of simulations reached agreement - and in virtually all successful cases, agents did so by adopting structured governance mechanisms such as voting rules, delegation, and agenda-setting. When governance was weak, even advanced models frequently failed. The key takeaway is that governance design shapes behavior more than model intelligence alone. For organizations deploying AI in procurement, compliance, or decision support, provider choice and institutional rules are strategic decisions, not technical details.

VCs evaluate AI startups. But what if the VC itself runs on AI? At Ellipsis Venture, we've built agentic systems for deal flow, diligence, and fund ops—turning a two-partner fund into something that operates like a much larger team. Moreover, if AI can help run a fund, can it also help founders build companies from ideation through validation to MVP? This talk is part demo, part manifesto. I'll show the tools we actually use, and present a framework for investors and founders building companies where AI isn't a feature but the foundation. If you're skeptical that agents can do real work, come and join the conversation.