Training
Dr. Yuki Tanaka
Research Lead, Post-training · Frontier Labs
RL from execution feedback at 10B tokens/day
Speakers
The people publishing the results, not the people summarising them.
Every speaker presents work they shipped or measured themselves. First wave announced — the full roster lands in October.
Open Weights
Amara Osei
Head of Open Weights · Mistral-scale collective
What the new open-weight frontier actually closed
Inference
Lars Møller
Principal Engineer, Inference · Tensor Runtime Co.
Speculative decoding and the cost curve of long-horizon agents
Agent Experience
Priya Raghunathan
Agent Experience Lead · Harness Systems
Agent experience: designing the surface humans actually supervise
Loop Engineering
Tomás Iglesias
Founder · Loopwork
Loop engineering: control flow for non-deterministic workers
Evals
Dr. Nadia Fahmy
Evaluation Scientist · Benchmark Institute
Contamination-resistant evals for agentic tasks
MCP
Sam Whitfield
Protocol Maintainer · MCP Working Group
MCP at scale: servers, sandboxes and capability negotiation
Harness
Chen Wei
Staff Engineer, Harness · Autonomy Systems
Harness engineering: sandboxes, retries and the cost of a bad tool
Benchmarks
Dr. Ines Duval
Benchmarks Lead · European AI Institute
Reading benchmark deltas without fooling yourself