Tag
LazyAgent introduces a demand-driven execution framework for agentic programs that selectively executes only necessary steps based on the current goal, leading to substantial efficiency gains and cost savings over eager baselines.
Presents LemonHarness, an integrated execution framework for long-horizon LLM agents that constrains state-changing operations within a clearly defined workspace, introduces a reusable rule knowledge base, and adds time-aware execution. Achieves 84-86% accuracy on Terminal-Bench 2.0.
PhoneHarness is a mixed-action benchmark and execution framework that evaluates phone-use agents on verifiable mobile workflows, achieving a 75% pass rate and outperforming existing approaches by 12.9 percentage points through deterministic action routing and auditable execution traces.