A feasibility experiment called qwen3-ebpf ran a full Qwen3-0.6B forward pass inside Linux eBPF, with BPF programs computing all 28 decoder layers, the output logits, and the next-token argmax while a…
A new empirical study from ActPlane analyzing 2,116 statements across 64 popular repositories finds that 64% of AI agent instructions are behavioral policies, but most require context and layered OS e…
AI infrastructure is shifting from token-based measurement to verifiable work units that bundle acceptance conditions, execution evidence, total cost, and accountability boundaries, as signaled by rec…
The SchedCP framework, detailed in a paper from arXiv (2509.01245), achieves up to 1.79x speedup and 13x lower search cost for Linux scheduler tuning by using a controlled optimization loop that separ…
A new open-source tool called agentpprof, part of the AgentSight project, uses semantic flamegraphs to aggregate AI agent trace data by intent, revealing where token budgets, time, and costs are conce…