Researchers have reanimated two landmark AI programs that modeled how people program, using executable archaeology to run their original code.
The first, Herbert Simon's Heuristic Compiler (1961), applied means-ends analysis from the General Problem Solver to model programming as problem-solving. The second, KBEmacs (1984) from MIT's Programmer's Apprentice project, let programmers build code by combining familiar "cliches" through a stylized English command language.
Both were reanimated via "reanimation" — running original code. The Heuristic Compiler reproduced compiled code from Simon's 1963 paper, while KBEmacs reproduced its December 1984 SQRT self-test. Side by side, both theories share common ground: the Heuristic Compiler also adapts existing routines, and Simon later argued search requires substantial programming knowledge.
To underscore the project's reflexivity, the team used Anthropic's Claude Code (Opus 5.5) for detailed reanimation work. These explicit, executable theories of programming are largely absent from today's more capable AI coding assistants.
Submission history: Jeff Shrager, v1, Mon, 26 Oct 2026 22:09:13 UTC (20 KB).
Source: Hacker News · Summarized by HeadlinesBriefing