Search papers, labs, and topics across Lattice.
This paper introduces StateAct, a novel multi-agent framework that leverages direct access to program state rather than relying solely on visual perception from screenshots. By employing a main agent that interacts with the underlying program state and a GUI subagent for limited visual tasks, StateAct significantly enhances task success rates while reducing operational costs. The results show a marked improvement in performance metrics on OSWorld 2.0, with binary success rates increasing from 20.6% to 26.9% and partial success rates from 54.8% to 61.6%, demonstrating that state-grounding shifts the focus from perception to reasoning in agent performance.
Shifting the focus from visual perception to direct program state access can boost agent success rates while slashing costs by nearly 9x.
Computer-use agents are usually improved by strengthening perception: better models for reading a screenshot and choosing where to click. Yet a screenshot is only a lossy rendering of the underlying program state, e.g., the files, application backends, and DOM that hold the task data. Different states can produce the same pixels, while code can inspect and modify that state directly. StateAct is a code-first, multi-agent harness built around this distinction. Its main agent works directly with program state by using code, while a dedicated GUI subagent handles screenshot-and-click interaction on the few subgoals that need it, just 28 of 108 tasks and 1.1% of main-agent steps. The same direct access to program state also supports verification: an independent finish gate double-checks the saved result for structural failures, e.g., output that is missing, unsaved, or written to the wrong path. To stay on track over hundreds of steps, the main agent hands subgoals to fresh subagents, keeping its own context focused. On OSWorld 2.0, StateAct lifts Claude Opus 4.8 from 20.6% to 26.9% on binary success, and from 54.8% to 61.6% on partial success, at ~ 9x lower cost per task than the same model driven by screenshots alone; a code-only variant with no GUI subagent reaches only 45.9% partial, below that screenshot-based baseline's 54.8%. In general, grounding action, verification, and memory in state, what we call state-grounding, shifts the main bottleneck from perception toward reasoning: failures depend more on what the agent thinks than on what it sees.
