If you havent recently used Claude Code, you might not understand where AI is at

A Hacker News thread is pointing readers to David Preichert’s Substack piece, “If you haven’t recently used Claude Code, you might not understand where AI is at,” with the discussion at item 49157496. At the time of the summary the post sat at 3 points and 1 comment, and the framing groups Claude Code with Codex and Fable as the reference set for how far coding-oriented AI has moved.

Claude Code is the product the author treats as the calibration tool: not a one-off chat demo but a coding agent people actually sit with in a real workflow. The argument is that older impressions of autocomplete or short Q&A no longer map to what these systems do when they stay in a project, take multi-step tasks, and work through the same kinds of edits and checks a developer would. Codex and Fable appear in the same conversation as peer or adjacent agents, so the claim is about a class of tools, not a single brand.

For engineers and builders, the practical gap is experience, not a press-release feature list. If the last serious try was months ago, product demos and benchmark screenshots will understate how much friction is gone from setup, iteration, and “finish this change” loops. That skews hiring, tooling budgets, and “should we adopt agents” debates toward yesterday’s failure modes.

Market-wise the piece sits in the Claude / Codex / Fable cluster that HN readers already associate with agentic coding, not generic chatbots. Early traction on the thread (3 points, 1 comment) is thin social proof; the signal is the title’s claim that recency of hands-on use is the main filter for whether someone’s mental model of AI coding is current.

Watch next is simple: use Claude Code (and, where relevant, Codex or Fable) on a real repo task, not a toy prompt, then reassess where agents help and where they still break. The post’s thesis is that without that refresh, commentary about “where AI is at” is lagging the product people are shipping with.