Co-Founder and CSO of Iris.ai, the expert AI harness that makes AI reliable enough for industries where a wrong answer isn't an option. Forbes Top 50 Women in Tech, two-time TEDx speaker and international keynote speaker who also built a race car and rides a Harley.

In this session, participants will learn to tell lasting shifts in AI from passing hype, using lessons from more than a decade of building an AI company.
Anita Schjøll Abildgaard co-founded Iris.ai at Singularity University in 2015, seven years before ChatGPT. Since then she has watched AI move from research labs into boardrooms, and she has seen investors, customers and the press lose interest in AI and come back again. Many of today's promises about AI were made before. Which of them held up? What did large language models change, and which problems are still as hard as they were in 2015?
Anita will take participants through Iris.ai's own journey, with its pivots, failures and successes. Iris.ai began with tools that helped researchers find scientific papers, with the goal of building an "AI researcher." Today it makes AI reliable for industrial companies. Anita will share what each phase taught her about data, trust and timing, and which of those lessons still apply. Participants will then use these patterns to look ahead and judge which of today's bets on AI are likely to pay off.
This session is ideal for leaders, investors and builders who make decisions about AI and want a longer view than the latest news cycle. Participants will leave with a practical filter for judging AI claims and a grounded sense of where the technology is going.
In this session, participants will learn what it takes to make AI reliable enough for work where a wrong answer isn't an option.
It is easy to make an AI demo look impressive. Production is different. In a steel plant, a car factory or a nuclear power station, an AI system that is right most of the time is not good enough. How can an organization know that an AI answer is right before anyone acts on it? And what will change when AI becomes reliable enough to work without a human checking every output?
Anita Schjøll Abildgaard will introduce AI harnessing: controlling what a model reads and measuring every step of how it answers. Drawing on Iris.ai's work with industrial companies, she will show where AI systems fail on expert technical documentation, and why a check of only the final answer misses many of those failures. Participants will see how Iris.ai evaluates the whole process with 23 metrics, from the instructions and prompts to the context the model received. The session will close with a look ahead at what becomes possible when industry can trust AI with its most critical knowledge.
This session is ideal for executives, CTOs and engineering leaders who are moving AI from pilot to production, especially in regulated or safety-critical industries. Participants will leave with clear questions to ask any AI vendor or internal team, and a higher standard for what "good enough" means.