AI Software Development
AI for the Future
I’ve been experimenting a lot with AI lately—specifically, AI-powered code editors. With tools like Windsurf and Cursor, the evolution of AI in development environments is becoming clearer. These aren’t just autocomplete tools anymore; they’re much more powerful, capable of scanning entire codebases, identifying issues, and applying fixes based on linting and type errors. They feel like having an extra set of hands to refine code, and so far, I’ve been impressed.
One of the most noticeable benefits of these tools is their ability to level the playing field. They help developers bridge gaps in unfamiliar parts of a codebase, making previously daunting tasks more accessible. This feels like a major shift—AI isn’t replacing developers; it’s augmenting them in ways that make learning and problem-solving more efficient.
I’ve been interested in experimenting with how people are approaching this. There are more and more examples popping up online and on social media of engineers using these tools to adopt paired programming practicing (themselves with the AI), or utilising AI within the code editor to provide a code review. There seems to be potential for editors like Cursor, Windsurf (and others) to be more than just code generation tools. Custom rulesets and prompts allow you to effectively create different personas (code reviewer, architect, fellow engineer) when interacting with the AI, and achieve different objectives.
The AI Divide: Mainstream vs. Niche Frameworks
One of the more frustrating aspects of using AI-generated code is the difference in quality between mainstream and niche frameworks. Popular languages like Python and JavaScript perform well because the AI has been trained on extensive data from sources like GitHub and Stack Overflow. But newer or more niche frameworks vary quite significantly in my experience.
Take Svelte 5 as an example. This blog is built with Svelte 5, and since its recent major release, I’ve noticed AI-generated code for it has been mixed. I’ve noticed a drop in quality of output with Svelte more generally vs Vue, React and more common languages. And due to the knowledge cutoff, this has naturally been exacerbated further by the major version release in Svelte. Despite careful prompting and custom instructions, I still need to regenerate responses occasionally to remind the AI that we’re working with Svelte 5. This is expected; models are only as good as their training data, and when a framework is relatively new or has undergone significant changes, there’s a lag in AI understanding.
This raises an interesting question: What happens to smaller, niche frameworks in an AI-dominated future? If AI-assisted development is shaping the way people learn and adopt technologies, and tools like AI editors become more ubiquitous with software development, will smaller tech struggle to gain traction simply because they aren’t well-supported by AI models?
Even within popular frameworks, this issue extends to frequent major releases. I’ve found similar challenges with NextJS 15 and AI insisting on older syntax and conventions. If AI does lag behind the present day, how much friction are these updates going to introduce to the developer experience?
The Role of AI Agents in Development
The move from simple AI suggestions to AI-driven agents represents a significant shift in development workflows. These agents don’t just assist—they act on your codebase, modifying files, suggesting changes, and running tests. This evolution of AI tooling makes me wonder where development is headed. There’s been a lot of focus on productivity and code generation. However, I am really curious about how these tools could change the day-to-day work for software engineers - planning, forecasting, understanding and evaluating. These tools are - and will continue to be - deeply integrated into the codebase, so the prospect that AI may more accurately be able assist in these areas feels highly likely to me. The only question is—how will this come about?
Final Thoughts
AI-powered development tools are evolving rapidly, and their impact is already being felt across different areas of software engineering. Whether it’s navigating legacy codebases, bridging knowledge gaps, or shaping the way frameworks are adopted, AI is becoming a crucial part of the development process.
However, the reliance on AI also introduces new challenges, particularly for smaller frameworks and frequently updated technologies. How we balance innovation with AI-assisted adoption remains an open question.
For now, I’ll continue experimenting with these tools and reflecting on their impact. AI isn’t replacing developers—it’s making us more effective. And as these tools improve, it will be interesting to see how development continues to evolve in response.