Software developers have watched AI coding tools evolve from advanced autocomplete to systems that can assemble full applications from a text prompt. Tools like Claude Code and Codex now participate in long-running projects, writing code, running tests, and, with supervision, fixing bugs. OpenAI has even integrated Codex into its own toolchain, publishing technical details about its inner workings.
Ars Technica spoke with several professional developers to gauge how these tools perform in practice. The responses show broad confidence that the technology works, but a division on whether that progress is entirely good news for the workforce.
David Hagerty, who develops point-of-sale software, cautions against marketing hype. “LLMs are revolutionary and will matter a lot, but they aren’t a magic substitute for human creativity,” he says, underscoring that the tools excel at structured coding tasks rather than novel storytelling.
Roland Dreier, a longtime Linux contributor, acknowledges hype but notes a genuine leap forward. He points to a recent improvement—spurred by Claude Opus 4.5—that turns AI from an autocomplete aid into a partner capable of debugging and delivering results. For complex tasks, he estimates as much as a tenfold speed boost, telling a Rust backend story that previously would have taken much longer.
Beyond performance, the articles describe a spectrum of views on job security and education. Some developers embrace the new paradigm, arguing that work will shift toward design, supervision, and architecture while AI handles repetitive coding. Others warn that widespread adoption could depress wages for junior roles or require retraining for the next generation of engineers.
In enterprise settings, the adoption gap becomes evident. Large organizations face overhead—legal reviews, data governance, and model access controls—that slow deployment, even as the same tools bolt onto widely used products and services. Nevertheless, a growing cohort of engineers reports that AI-assisted workflows can dramatically accelerate feature delivery, prototype testing, and modernization of aging codebases, albeit with caution about potential debt and hallucinations.
