Appshots
person concept tool org talk claim — click a node to jump to its page; hover an arrow for the relation
Appshots is a tool feature that captures both a visual screenshot and the full accessibility tree of an application, providing richer context to AI agents than a standard screenshot alone.
What Appshots Does
Unlike a plain screenshot, Appshots records the entire accessibility tree of an app — including element IDs and structural metadata — alongside the image. According to Jason Liu, this additional structured information enables AI agents to make single-hop tool calls rather than requiring multi-step interactions to locate and identify UI elements. As Liu explains: "with appshots, it takes not only the image but the entire accessibility tree of the app," making it possible for an agent to act on a specific element directly because it already has the ID in context. 16:25
Advocacy
Jason Liu (OpenAI Codex workshop) describes Appshots as "my favorite feature of all time," positioning it as a significant quality-of-life improvement for anyone building or using AI agents that interact with desktop or application interfaces. 0:28 16:25
The core argument for Appshots over plain screenshots is efficiency and precision: because the accessibility tree exposes actionable element IDs, the AI does not need to visually infer or search for a target element — it can reference it directly, reducing latency and error in agentic UI workflows.