Mistral AI, a French startup backed by Microsoft, has introduced Codestral, its first generative AI coding assistant, which supports over 80 programming languages, including mainstream languages such as Java, Python, C++, JavaScript, and more niche ones like Swift, and Fortran. Codestral is entering a competitive market dominated by platforms like GitHub Copilot and Code Llama.
That’s the headline. The question you have to ask: “What will it mean to be a software engineer in 60 months?” This specific use case for LLMs is progressing at a remarkable pace. So fast, in fact, that it is sparking vigorous debate about the adage, “Teach your kids to code.”
Back to Codestral: Mistral AI says Codestral (which operates on a 22B parameter model) sets new standards for performance and latency in code generation. Unlike other Mistral models, Codestral is not fully open-source. Instead, it is described as an “open-weight” model under the Mistral AI Non-Production License, restricting its use to research and testing purposes only and prohibiting commercial use. Mistral offers a hosted version of Codestral on its Le Chat conversational AI platform and a paid API. The company is planning to integrate Codestral into development environments such as LlamaIndex, LangChain, Continue.dev, and Tabnine.
AI coding assistants are very popular. It isn’t hard to understand why. A study by GitHub indicated that developers using AI assistants completed tasks 55% faster. That’s the good news. The bad news? Research from GitClear revealed increased code churn and concerns about reliability and accuracy in code generated by AI assistants. It gets worse, as AI-generated code often amplifies existing bugs and security vulnerabilities. The icing on the cake? A study from Purdue University found that more than half of the answers provided by OpenAI’s ChatGPT to programming questions were incorrect.
Despite these challenges, the market for AI coding assistants is growing, which will undoubtedly fuel ongoing debates about the benefits and risks of relying on AI for software development, as well as force you to consider how engineering teams will evolve over the next few years.
Author’s note: This is not a sponsored post. I am the author of this article and it expresses my own opinions. I am not, nor is my company, receiving compensation for it. This work was created with the assistance of various generative AI models.