AI productivity tips for migration projects

I’ll try to comment myself. Maybe it’ll break the ice, maybe not.

1. I use Cursor. I really like the level of integration it has with AI tools and how they can be used in a synergistic way. For migration projects, the most useful features are: Workspaces, File referencing, Documentation indexing, and URL referencing. These features exist in other AI code editors as well, and I try those regularly to see how they evolve, but Cursor is still the winner for me.
2. For code migration, the agentic approach is the way forward. In a workspace, I open both the source and target solutions side by side. To migrate a specific piece of functionality, I reference all relevant files from the source solution in my prompt, along with additional context like links to Xperience by Kentico documentation. Then the agent generates the migrated code in the target solution and does its best to adjust it to the target’s requirements. From there, I can continue refining it.
3. I’ve been struggling more with complex data migrations, where heavy adjustments to the target data and structure are needed. Writing the first few migration scripts was intense, but I found that once you have a couple of them done, Cursor can take inspiration from those to create new ones. And since these scripts are one-off (non-production code), leaning into a bit of “vibe coding” with proper reviews and testing actually turns out to be quite productive.

I'm happy to join in, with the caveat that I'm not working on any upgrades. However, I can share my thoughts based on my experiences doing upgrades when I worked for a partner agency.

1. VS Code with GitHub Copilot + some MCP servers (like Context7).
   I tried Cursor and absolutely loved some things about it, especially when I'm focused on getting the AI agent to do as much as possible - it is very good at managing context. I'm not thrilled about some of the UX of the editor and Cursor currently has issues with PowerShell (heavily used in the Kentico Community Portal). Once .NET 10 comes out and I can replace the PowerShell scripts with file-based apps, I'm going to revisit Cursor.
2. I was already using 2 editors every time I worked on a KX13 project - VS Code for the front-end and ASP.NET core app, and VS for the classic ASP.NET app. It was annoying but I got used to it and it let me use the best, modern developer tools in VS Code as much as possible. That said, now I would bring the KX13 ASP.NET Core project and its .NET Standard libraries into the XbyK solution. The AI agent doesn't need access to the web forms application.
3. I imagine the most difficult thing is managing context - especially when the agent has to understand the similarities and differences between KX13 and XbyK architecture and APIs. In VS Code with Copilot I'd try to solve this with clear instruction files and maybe some custom chat modes that automatically set the context for the agent.
   If you are keeping the same front-end experience, then Playwright end-to-end tests, generated by AI running against the KX13 project and then tested against the XbyK project could help give you confidence that you're not missing important features or introducing regressions.