Software program builders have spent the previous two years watching AI coding instruments evolve from superior autocomplete into one thing that may, in some cases, construct total functions from a textual content immediate. Instruments like Anthropic’s Claude Code and OpenAI’s Codex can now work on software program tasks for hours at a time, writing code, operating checks, and, with human supervision, fixing bugs. OpenAI says it now makes use of Codex to construct Codex itself, and the corporate not too long ago published technical particulars about how the device works beneath the hood. It has induced many to marvel: Is that this simply extra AI business hype, or are issues really completely different this time?
To search out out, Ars reached out to a number of skilled builders on Bluesky to ask how they really feel about these instruments in follow, and the responses revealed a workforce that largely agrees the know-how works, however stays divided on whether or not that’s fully excellent news. It’s a small pattern measurement that was self-selected by those that wished to take part, however their views are nonetheless instructive as working professionals within the area.
David Hagerty, a developer who works on point-of-sale methods, informed Ars Technica up entrance that he’s skeptical of the advertising. “All the AI firms are hyping up the capabilities a lot,” he stated. “Don’t get me improper—LLMs are revolutionary and may have an immense influence, however don’t anticipate them to ever write the following nice American novel or something. It’s not how they work.”
Roland Dreier, a software program engineer who has contributed extensively to the Linux kernel prior to now, informed Ars Technica that he acknowledges the presence of hype however has watched the development of the AI area intently. “It feels like implausible hype, however state-of-the-art brokers are simply staggeringly good proper now,” he stated. Dreier described a “step-change” prior to now six months, notably after Anthropic launched Claude Opus 4.5. The place he as soon as used AI for autocomplete and asking the occasional query, he now expects to inform an agent “this check is failing, debug it and repair it for me” and have it work. He estimated a 10x velocity enchancment for advanced duties like constructing a Rust backend service with Terraform deployment configuration and a Svelte frontend.
