agentic-ai-summit
From ChatGPT to OpenClaw: How Agentic AI is Redefining AI Compute and the Open Evaluation
The evolution from ChatGPT's conversational AI to OpenClaw's autonomous agent ecosystem marks a paradigm shift in AI compute. This transformation is fundamentally bidirectional: while advanced compute infrastructure enables increasingly sophisticated agentic capabilities, the agentic paradigm itself is reshaping compute delivery through modular, skill-based architectures.
OpenClaw exemplifies this new wave, where agentic AI system capabilities are determined by the interplay of large models, skills, and agent systems. This tripartite architecture demands new evaluation methodologies that assess not just model performance, but the orchestration of skills and autonomous decision-making—addressed by open evaluation initiatives like FlagEval and the Eclipse partnership PanEval.
FlagOS sits at the center of this transformation, building universal and inclusive AI compute capabilities that support heterogeneous hardware backends. By enabling efficient execution of the massive token volumes demanded by agentic workflows, FlagOS provides the computational substrate for this new era—where compute is not merely consumed by intelligence, but actively shaped by it.
In this talk, we will share insights on how to architect AI infrastructure for the agentic era, including the latest evaluation methodologies for agentic AI systems that go beyond traditional benchmarks to assess skill orchestration and autonomous decision-making. We will demonstrate how FlagOS Open Compute enables bidirectional empowerment: powering OpenClaw's universally accessible capabilities through cross-chip unified deployment, while OpenClaw agents feed back into FlagOS via automated operator generation and model migration. We will also discuss the path toward open, transparent evaluation standards that ensure trustworthy autonomous systems.
OpenClaw exemplifies this new wave, where agentic AI system capabilities are determined by the interplay of large models, skills, and agent systems. This tripartite architecture demands new evaluation methodologies that assess not just model performance, but the orchestration of skills and autonomous decision-making—addressed by open evaluation initiatives like FlagEval and the Eclipse partnership PanEval.
FlagOS sits at the center of this transformation, building universal and inclusive AI compute capabilities that support heterogeneous hardware backends. By enabling efficient execution of the massive token volumes demanded by agentic workflows, FlagOS provides the computational substrate for this new era—where compute is not merely consumed by intelligence, but actively shaped by it.
In this talk, we will share insights on how to architect AI infrastructure for the agentic era, including the latest evaluation methodologies for agentic AI systems that go beyond traditional benchmarks to assess skill orchestration and autonomous decision-making. We will demonstrate how FlagOS Open Compute enables bidirectional empowerment: powering OpenClaw's universally accessible capabilities through cross-chip unified deployment, while OpenClaw agents feed back into FlagOS via automated operator generation and model migration. We will also discuss the path toward open, transparent evaluation standards that ensure trustworthy autonomous systems.
Intervenants
