Geoffrey Litt is a software designer and HCI researcher whose career circles one idea: malleable software — computing environments where anyone can adapt their tools to their own needs with minimal friction. He is currently a design engineer at Notion. Before that he spent years at the independent research lab Ink & Switch, where he helped articulate and prototype the malleable-software vision, and before that he did a PhD in human-computer interaction at MIT's CSAIL Software Design Group, advised by Daniel Jackson.

## From startups to a research practice

Litt studied electrical engineering and computer science at Yale (class of 2014, with distinction in the major, per his resume) and joined Panorama Education — a Boston startup building data tools for K-12 schools — as one of its first engineers. His resume describes him as the third engineering hire; he designed the company's data reporting interface and led an NLP feature that categorized open-ended survey feedback for educators. On the Dialectic podcast he has described this period as formative: teachers kept asking for small changes the centralized team could not ship, which seeded his obsession with letting users adapt software themselves. He attended the Recurse Center in 2018, then began a PhD at MIT in September 2019.

His 2023 dissertation, *Building Personal Software with Reactive Databases*, extends spreadsheet and relational-database techniques through three systems. Wildcard (2020) is a browser extension for spreadsheet-driven customization: it scrapes a web app's underlying data into a table view so end users can sort, annotate, and extend real sites without traditional programming. Potluck and Riffle followed — the other two pillars of the thesis.

## The Ink & Switch years

Ink & Switch is an independent research lab — in the Dialectic description, one "focused on how computers can help us think and work" — that builds prototypes exploring new computing platforms. Litt collaborated with the lab during his graduate summers and joined full-time after finishing his doctorate, eventually as a senior researcher leading malleable-software work.

His projects there fall into two braided strands. The local-first strand produced Peritext (with Sarah Lim, Martin Kleppmann, and Peter van Hardenberg), an intent-preserving CRDT for rich-text collaboration published in PACM HCI at CSCW 2022; contributions to Cambria, which translates data between apps via bidirectional lenses; and Patchwork, a universal version control project he started in January 2024 with Paul Sonnentag, Max Schöning, Adam Wiggins, van Hardenberg, and Orion Henry. Patchwork prototyped branching, diffing, and commenting primitives inside the lab's Automerge-based writing tool — the team wrote its own public lab notebook inside the prototype — and asked whether version control could someday live in the OS storage layer.

The dynamic-documents strand produced Potluck (2022, with Schoening, Paul Shen, and Sonnentag), which lets users gradually enrich text notes into interactive tools through live searches, formulas, and dynamic annotations; and Embark (2023, with Sonnentag and Alexander Obenauer), which applied the same model to travel planning. Both were presented at the LIVE workshop on live programming, and both reflect a signature method: prototypes tested in serious contexts of authentic use, then written up as public essays rather than shipped as products.

The strands converge in his writing. In June 2025 Ink & Switch published *Malleable software: Restoring user agency in a world of locked-down apps*, co-authored by Litt, Josh Horowitz, van Hardenberg, and Todd Matthews — a manifesto arguing that mass-produced apps are too rigid, that modification should be routine at the point of use, and that users deserve a gentle slope from using software to creating it, drawing on HyperCard, Smalltalk, and Dynamicland.

## AI and the move to Notion

Since early 2023 Litt has written about how LLMs change the equation for personal software. In *Malleable software in the age of LLMs* he argued that models can handle the fuzzy parts — extracting structure from messy text — while users keep control of the actual computations; he has been consistent that AI coding alone does not deliver malleability. His 2025 writing got more practical: *Stevens*, a hackable personal assistant built from one SQLite table and cron jobs; *Enough AI copilots! We need AI HUDs*; and *Code like a surgeon*, a workflow that delegates secondary tasks to background agents while keeping the core craft hands-on.

In September 2025 he left Ink & Switch and joined Notion as a design engineer, reuniting with former collaborator Max Schoening. His stated rationale: Notion already embodies "document to app" thinking, is a ready substrate for human–AI co-creation, and has a distribution model — templates shared inside companies and communities — suited to personal software. He has called it the company best positioned to democratize software creation in the age of AI. What he is building there has not been publicly detailed; a Recurse Center talk description mentions early experiments at Notion alongside the HyperCard-to-Embark arc.

Alongside the writing he has kept up a steady interview circuit — Metamuse on bring-your-own-client, localfirst.fm on Riffle and malleable software, Dialectic's long "Software You Can Shape" conversation, Dive Club, and Imbue's Generally Intelligent — which together make his stated philosophy unusually well documented for a working researcher.

## What the record does not settle

The skeleton of this biography — Yale, Panorama, the PhD timeline, the Ink & Switch tenure — rests largely on Litt's own resume, site, newsletter, and self-maintained profiles; independent reporting on him personally is thin. His Wikidata entry is essentially an ORCID stub and there is no Wikipedia article. The status of Patchwork post-departure, and what his design-engineer work at Notion will produce, are open questions the sources pose but do not answer.

*This index was compiled from public sources and does not imply the subject's endorsement. Citations live in the packet's source catalog.*