Ankur Sethi
Software developer who manually retypes LLM-proposed edits to preserve comprehension, taste, and a spatial mental model of his codebase.
Ankur Sethi is a software developer and writer whose Lab Notebook covers programming, personal projects, and the experience of using coding assistants without surrendering comprehension. In “Prevent cognitive debt by manually retyping LLM-generated code,” he describes a deliberately slow workflow: the model proposes edits in chat, while he manually types, adapts, and verifies every change.1
Sethi treats manual retyping not as nostalgia for typing but as an active review surface. The physical sequence of entering each change gives him time to notice hallucinations and questionable design, refactor toward his own taste, stop on unfamiliar APIs, and preserve a spatial mental map of the codebase. He estimates that this gives him roughly a 2× rather than 10× speedup, trading maximum throughput for deeper ownership.1
His workflow is a concrete mitigation for Cognitive Surrender and an applied version of Code Reading as Cognitive Labor. It also offers a deliberately conservative mode of AI-Assisted Software Development for personal projects where learning, enjoyment, and long-term comprehension matter more than raw delivery speed.
In the blog
SummaryType it yourself: a deliberately slow workflow for AI-generated code
Mentioned in 3 pages
Thinking next to AI
Engineering with agents