LLMs as Developer Tools: When Augmentation Becomes the Norm

LLMs are no longer just chatty assistants; developers are using them as copilots that augment toolchains. A post about Scheme Machine argues these models should aid programming tools like grammars and compilers, not replace programmers, keeping the focus on building robust workflows ^[1].

Inside VS Code with the Hugging Face Copilot Chat extension, open models like Qwen3, GLM 4.6, and Kimi K2 0905 are delivering solid refactoring and quick edits. They’re cheaper, avoid vendor lock-in, and you can swap models on the fly. Earlier, the author used Cursor for a smoother AI coding experience, showing how open tooling can fit existing IDEs ^[3].

Model-selection chatter in October 2025 shows a tug-of-war between open weights and proprietary engines. Open-weight stars Kimi K2 0905, Qwen 3 Coder 485B, and GPT5 Mini are highlighted, with GPT5 Mini billed as cheapest. One post even quotes Kimi K2 0905: "writes better than 95% of humans". Some discussions point to Qwen Code CLI boosting coding with open weights, underscoring open models’ growing playground in real work ^[2].

On the ideological front, the Scheme Machine argument meshes with practical bootstrapping chatter: GNU Mes and Guix folks are pursuing bootstrap paths—getting a Scheme environment to run a C compiler that can bootstrap from a tiny bootloader—toward a more AI-assisted toolchain ^[1].

The throughline is clear: AI in development is about augmentation and orchestration, not replacement. Expect dev workflows to feel more like a coordinated stack of AI-assisted tools than a single model doing all the heavy lifting.

References