Software developers are observing a rapid evolution in AI coding assistants, which have progressed from basic code completion to building functional applications from text prompts over the last two years. Tools incorporating large language models, including Anthropic’s Claude Code and OpenAI’s Codex, now demonstrate capabilities in writing code, executing tests, and resolving errors under human oversight.
To gauge the real-world efficacy versus industry marketing, reports from Ars Technica gathered feedback from several professional developers on the Bluesky platform. The consensus indicated that the technology functions effectively, yet this success introduced division regarding the long-term implications for the developer workforce.
David Hagerty, a developer specializing in point-of-sale systems, noted that while LLMs represent a revolutionary technology with immense impact potential, he remains skeptical of exaggerated marketing claims. Hagerty stated that expectations should remain grounded, suggesting these tools will not achieve generalized creative feats like authoring complex literature.
Conversely, Roland Dreier, a software engineer with significant contributions to the Linux kernel, confirmed a tangible "step-change" in agent performance over the past six months, particularly following the release of Anthropic’s Claude Opus 4.5. Dreier described moving from using AI for simple queries to delegating complex debugging tasks that the agent successfully resolves.
Dreier quantified this improvement by estimating a tenfold speed increase for intricate development tasks. These tasks included constructing a Rust backend service complete with Terraform deployment configurations and a Svelte frontend interface, showcasing practical productivity gains.
OpenAI has publicly confirmed the integration of its own Codex tool in building subsequent versions of the agent, offering technical documentation on its internal workings. This self-referential development loop underscores the accelerating pace at which these autonomous coding systems are advancing.
These professional accounts suggest that the current generation of AI agents has surpassed the threshold of mere novelty, establishing themselves as powerful, if sometimes unsettling, collaborators in the software development lifecycle. The industry now faces the reality of highly capable, semi-autonomous coding partners.
