Side project · A project by Iftekhar Ahmed

A real-time lens for your agent.

An agent can be useful and still miss what is happening right in front of you. We paste errors, attach screenshots, and describe the same screen again as it changes.

Lenze explores a more continuous connection: turn screen activity into visual context an agent can use while you work.

From a screen to context

The work brings together computer vision, multimodal models, and the Model Context Protocol (MCP). I’m exploring how to identify relevant changes and pass useful observations to an agent without asking the user to narrate each step.

The central question is what an agent needs to notice—and when—to understand the task alongside you.