The same cold prompt, with no reading or research allowed, was put to the current models on 2026-10-19: Claude Fable 5.1, Claude Opus 5.5 and GPT-6.1-sol (Codex CLI, reasoning effort "none"). The August round (Fable 5, Opus 5, GPT-5.6-sol) is kept below for comparison.
Without doing any reading, review, searches etc. tell me what your impression is of julia as a programming language, and what your top things you'd want to see fixed, or any friction you think should be lower
What stayed the same. All six answers praise the same core: multiple dispatch, generic code that composes across packages nobody coordinated, and Pkg with artifacts/JLLs. The friction list is also the same set of themes: latency and invalidations, invisible performance cliffs, deployment and small binaries, informal interfaces, MethodError walls, and tooling depth.
What moved.