Are We at the Tipping Point for Agentic AI Deployment?

This session, co-organized by the ELLIS Human-centric Machine Learning Program and ELLIOT, is part of the ELLIS UnConference 2026, held at the Palais des Congrès de Paris on the 8th of December 2026. The session is a two-hour structured debate on whether agentic AI is ready to be deployed at scale.

ELLIS UnConference

The ELLIS UnConference is an annual, one-day gathering of the European machine learning community focused on exchange, collaboration, and scientific discussion. The 2026 edition kicks off the NeurIPS satellite event in Paris and features keynotes, parallel two-hour sessions, and a poster session.

Are We at the Tipping Point for Agentic AI Deployment?

As agentic AI moves from experimental demonstrations toward broader deployment, the machine learning community faces a pressing question: which claims about the capabilities and consequences of increasingly autonomous systems are supported by technical evidence and what are the associated risks? Recent developments give this question particular urgency, including the OpenAI–Hugging Face “incident”, OpenAI’s announcement of an AI-generated solution to the Navier–Stokes Millennium Prize Problem, and Amodei’s call for slower capability development and stronger independent oversight. Together, these developments have intensified debate about both the scientific promise and the risks of agentic AI. Yet moving from particular achievements or failures to broader predictions requires assumptions that deserve rigorous, technical and independent scrutiny. Public debate is shaped by a relatively small number of influential voices, operating within commercial, economic, and geopolitical contexts. These perspectives deserve consideration, but the research community has a responsibility to assess the underlying technical arguments independently. Doing so requires space for substantive disagreement, where both ambitious predictions and sceptical responses can be examined against evidence.

The session will examine what current research implies for the capabilities, risks, and deployment of agentic AI. We bring these question to the community through a structured debate: two short talks arguing opposite sides and a technical Q&A, followed by small-group discussions in which participants build and present their own arguments for and against.

The session is intended for machine learning researchers interested in agent capabilities, evaluation, safety, and deployment. Its aim is to identify where evidence supports concern, where uncertainty remains, and which claims can be translated into concrete research questions. Participants should leave with a clearer understanding of the technical disagreements and the experiments, evaluations, or theoretical advances needed to resolve them.

The session is organized by Nuria Oliver, Novi Quadrianto, Adrian Weller, Plamen P. Angelov, and Javier Sanguino.

Schedule

This session will take place on the 8th of December, 2026, at the Palais des Congrès de Paris. Times are in CET (UTC+1).

Registration

Registration for the ELLIS UnConference is open via the official registration form. See the ELLIS UnConference website for more details.

Venue

The ELLIS UnConference will be held at the Palais des Congrès de Paris, 2 Place de la Porte Maillot, 75017 Paris, France. The room for this session will be announced closer to the event.