Geoffrey Litt

Geoffrey Litt is a design engineer at Notion who builds malleable software: computing environments people can adapt to their own needs. His research asks how AI can make software easier to reshape while helping its users understand what they create.

Early work at an education-technology startup exposed him to schools struggling with standardized software that could not accommodate their particular requirements. He subsequently earned a PhD in human-computer interaction at MIT, advised by Daniel Jackson, and became a senior researcher at Ink & Switch.

Litt’s projects consistently make software behavior visible. JSON Sheets, developed in 2019, lets programmers observe data transformations as they happen. His 2020 TodoMVC Vis organizes a Redux application’s activity into a navigable debugging timeline. Ladybug connects a custom Ruby debugger to Chrome DevTools, bringing familiar browser-based inspection tools to backend code. At Dynamicland, he experimented with adaptable programs in an existing collaborative computing environment built around physical space and movable paper.

In September 2023, Litt began leading Ink & Switch’s malleable-software research track. With Paul Sonnentag and Alexander Obenauer, he developed Embark, a system that enriches travel notes with maps, routes, weather, and dates without requiring users to abandon their original document. He also contributed to Patchwork, which explores editable software and version control for collaborative creative work.

His 2025 framework for malleable software, coauthored with Josh Horowitz, Peter van Hardenberg, and Todd Matthews, advances three principles: gradual transitions from using software to changing it, interoperable tools connected through shared data, and communities capable of exchanging improvements. Litt joined Notion in September 2025, applying these ideas to collaborative documents and AI-assisted workflows.

• Documents that become applications. Embark and Litt’s experiments turning AI-assisted workflows into graphical interfaces begin with familiar notes or processes, then add interactive views and structured information as requirements emerge.

• Understanding as the new bottleneck. Litt argues that effective coding agents reduce the burden of writing and verifying code while creating a different risk: people lose the mental models needed to direct projects and generate their next ideas. His account of cognitive debt and creative participation distinguishes shipping correct software from maintaining genuine human comprehension.

• Explain Diff. His workflow transforms agent-generated changes into teaching materials: relevant background, conceptual explanations, interactive illustrations, and guided code walkthroughs. A short comprehension quiz checks whether the programmer understands a change before submitting it for team review.

• Agent-generated microworlds. Inspired by Seymour Papert’s educational approach, Litt creates temporary interfaces that make complex systems explorable, including a step-by-step Prolog interpreter debugger and an interactive website-migration tool. These environments build intuition about processes that automated execution can conceal.

Litt extends that emphasis on comprehension to teams through shared explainers and collaborative conversations involving people and coding agents. His call for agents optimized for understandable communication captures his larger conviction: powerful software should strengthen human agency, not leave its operators guessing.