-
Erwin Laure (MPCDF)07/10/2026, 09:00
-
Jakob Roth (MPCDF), Nastassya Horlava (MPCDF), Piero Coronica (MGRZ)07/10/2026, 09:20
This hands-on tutorial takes AI agents apart into their building blocks: the LLM "brain", tools, memory and planning. Participants build a literature-digest agent step by step in Jupyter notebooks, using the smolagents framework and an open-weight LLM served on MPCDF's LLM Inference Service. They start by talking to an LLM directly, then give the agent tools (hand-written and via the Model...
Go to contribution page -
Jakob Roth (MPCDF), Nastassya Horlava (MPCDF), Piero Coronica (MGRZ)07/10/2026, 11:00
This hands-on tutorial takes AI agents apart into their building blocks: the LLM "brain", tools, memory and planning. Participants build a literature-digest agent step by step in Jupyter notebooks, using the smolagents framework and an open-weight LLM served on MPCDF's LLM Inference Service. They start by talking to an LLM directly, then give the agent tools (hand-written and via the Model...
Go to contribution page -
Klaus Reuter (MPCDF)07/10/2026, 11:45
-
Dr Heiko Appel (MPSD), Dr Nicolas Tancogne-Dejean (MPSD)07/10/2026, 13:30
Within the last years, core developers of the Octopus code
Go to contribution page
(https://octopus-code.org) have started to integrate AI coding agents
into their daily workflows. In this talk, we present several aspects
that have been explored. In particular, we focus on two main directions:
one consists of using AI agents to consolidate and improve the existing
code base, and the second one is to implement... -
Dr Jan Janssen (MPIE), Prof. Jörg Neugebauer (MPIE)07/10/2026, 14:30
Modern materials simulations are increasingly evolving from predefined sequences of calculations toward adaptive computational processes in which subsequent steps depend on intermediate results. High-throughput screening, convergence studies, and multiscale simulations therefore require adaptive frameworks that can autonomously explore parameter spaces, assess results, and adapt computational...
Go to contribution page -
Dr Juan Manuel Lombardi (FHI), Dr Maurits Willem Vuijk (FHI), Dr Simeon Beinlich (FHI)07/10/2026, 16:00
At FHI, agentic AI is being explored across a range of applications spanning computational and experimental research.
Go to contribution page
We present three perspectives: an on-premise agentic AI platform supporting general-purpose assistance for data analysis, coding, and literature research;
a multi-agent system for autonomous exploration of atomistic configuration spaces;
and an agentic control loop combining... -
Oliver Beye (CBS)07/10/2026, 17:00
Artificial intelligence opens up new possibilities – and sometimes accelerates our thoughts to such an extent that we can barely keep up. At the same time, new obligations arise: data protection, the AI Regulation, good scientific practice. To ensure that good ideas do not fall by the wayside, what is needed is not a list of prohibitions, but reliable guidance: what is permitted and under what...
Go to contribution page -
08/10/2026, 09:00
-
David Carmona (Microsoft)08/10/2026, 09:10
Agentic AI is beginning to change how scientific discovery is done - moving from tools that answer a single prompt to assistants that can plan, reason across sources, and carry a research task through multiple steps. This session introduces what agentic AI means for the Max Planck research ecosystem: where it can genuinely accelerate discovery, how it can extract and connect knowledge across...
Go to contribution page -
Dr Mara Kiefer (Microsoft (GitHub Next)), Taja Ly (Microsoft)08/10/2026, 10:40
Format: Keynote (10 min presentation + 5 min questions) + Interactive Workshop (75 min), 90 min total
Abstract:
Go to contribution page
Imagine an always-ready engineering system where specialized AI agents discover valuable work, keep hundreds of repositories healthy, and safely deliver improvements while teams focus on what matters most. This is the promise of agentic campaigns: coordinated sets of... -
Elmar Husmann (AWS), Mónica Cortés (AWS), Tobias Leemann (AWS)08/10/2026, 13:30
AWS presents their latest solutions and services for AI-assisted research workflows
Go to contribution page -
Luka Stanisic (AMD)08/10/2026, 15:30
Agentic AI is shifting from assistants that answer a single prompt to systems that plan, call tools, and carry a scientific workflow through to completion. This talk grounds that shift in open hardware and open software: what separates an agent from the LLM underneath it, what actually governs agent reliability, and where agents already deliver for science — including a four-agent inertial...
Go to contribution page -
Marcelo Sanchez (NVIDIA)08/10/2026, 16:15
Agentic AI is beginning to change how complex workflows are executed—moving from tools that answer a single prompt to always-on autonomous assistants that can plan, reason across systems, and carry a task through multiple steps. This session introduces what agentic AI means when we prioritize safety and governance: where autonomous agents can genuinely accelerate productivity, how we can...
Go to contribution page -
Andres Aldana (Anthropic)08/10/2026, 17:15
This talk introduces Claude Science, Anthropic's AI workbench for scientific research: what it does and how it works. Two walkthroughs then follow a research question through to a result the scientist can check, with visible steps, code and sources. It closes with recent research examples carried out with Claude Science.
Go to contribution page
Choose timezone
Your profile timezone: