New language 'Jacquard' imagines code AI would prefer to write

New language 'Jacquard' imagines code AI would prefer to write

Jacquard is an experiment in what a might look like if it were designed by AI, for AI, rather than for humans. Its creator had AI analyze the abstract syntax trees (AST, a tree-shaped representation of code's structure) of several mainstream languages plus a few unusual but conceptually important ones, then used that analysis to build a new syntax and structure.

The name comes from the , an early automation device that used punch cards to weave patterns and is considered a precursor to Babbage's Analytical Engine. In this language, interactions with the outside world — like touching the filesystem or network — must appear explicitly in a function's signature (the definition of its name and inputs/outputs), and the runtime requires explicit permission before performing them.

Side effects during execution can be recorded and replayed later under different conditions. Code is also given a content-based identity, so renaming variables or reformatting doesn't force a recompile or retest.

Key points

  • AI was used to analyze the ASTs of multiple languages to help design the new syntax
  • External interactions (filesystem, network) must show up explicitly in s
  • Side effects can be recorded and replayed under different conditions
  • Content-based code identity means renames/formatting don't require recompiling or retesting
  • Named after the ; built with AI-written, human-reviewed code in mind
Read original