01
Open-weight models
Release-by-release analysis of what open weights can now do, licensing reality, and where the gap with closed frontier models actually sits.
Program
8 tracks. 2 days. Running in parallel.
Open weight models vs Frontier Model : defining the future of API
The next decade of AI infrastructure will be shaped by a single architectural bet: will applications call closed frontier APIs, or will they run open-weight models behind their own endpoints? Generation AI 2026 puts that question at the center — across every track, every talk, and every hallway conversation.
02
Benchmarks & evals
Contamination, saturation, agentic task suites, and how to build internal evals that survive a model swap.
03
Training & post-training
Pretraining data curation, RL from execution feedback, distillation, and fine-tuning that still pays for itself.
04
Inference
Serving economics, KV cache strategy, speculative decoding, and latency budgets for multi-step agents.
05
Agent experience
The human surface of autonomous work: approvals, interruption, legibility, and trust calibration.
06
Loop & harness engineering
Control flow, retries, context compaction, sandboxing and tool design for long-running agent loops.
07
MCP & interop
Model Context Protocol servers in production: capability negotiation, auth, and the emerging tool ecosystem.
08
AI infrastructure & systems
Production pipelines, observability, cost governance, and the platform work that lets agent teams move fast without breaking the budget.