geoffreylitt-com

This essay redirects to Ink & Switch's article on malleable software, which explores restoring user agency in a world of locked-down apps.

In a July 2026 talk at the AI Engineer conference, Geoffrey Litt, a Design Engineer at Notion, argues that understanding code remains critically important even when using AI coding agents, presenting the idea that comprehension—not generation—has become the new bottleneck in software development.

OpenAI released GPT-4 with impressive coding capabilities, while Microsoft Research published a paper showing GPT-4's sophisticated abilities. The article discusses how large language models are enabling new forms of end-user programming and malleable software.

The article discusses how large language models can enable users to create personalized software tools through flexible graphical interfaces. It builds on previous work exploring how LLMs might usher in a new era of end-user programming and malleable software development.

The author reflects on year-end book reading announcements but notes they don't track their own reading count. Instead, they focus on books that made a lasting impression throughout 2023.

The article discusses how people often feel pressure to create something completely original, using the analogy of a peach pie. It suggests that not everything needs to be groundbreakingly unique to have value.

The author describes using AI to build a custom debugger UI that made programming more enjoyable, without having the AI write any code. This approach enhanced the debugging experience through personalized tooling.

The article discusses the concept of avoiding the "nightmare bicycle" in product design, which refers to overcomplicating products with unnecessary features and labels instead of intuitive, simple interfaces.

Geoffrey Litt describes creating a hackable AI assistant called Stevens using just a single SQLite table and cron jobs. The system demonstrates that useful personal AI tools can be built without complex techniques or libraries.

The article presents a Socratic dialogue exploring whether chat is an effective user interface for AI. It examines the tension between design critiques of chat interfaces and their widespread adoption in AI products.

The article critiques the "copilot" metaphor for AI, referencing a 1992 talk by researcher Mark Weiser. It argues that instead of AI copilots, we need AI HUDs (heads-up displays) as a better design approach for human-AI interaction.

The article presents a thought experiment comparing AI's societal effects to teleportation. It imagines a future where an "Auto Go Instant" teleporter allows people to go anywhere instantly, exploring the implications of such technology.

The author argues that viewing AI as making everyone "managers" or "editors" is incomplete. Instead, they advocate for coding like a surgeon who does actual work while leveraging their highly skilled expertise and time effectively.