Blog Discussion: AI-Powered Refactoring: Updating Xperience by Kentico Logging with a Single Prompt

It’s quite interesting to see how different AI code editors can lead you to take completely different approaches to the same task.

I tried performing the same API update in Cursor. In the prompt window, I referenced the files where IEventLogService was used, then added links to the Logging documentation page, the Changelog section “Event log changes and recommendations,” and your article. After that, I asked it to perform the API update, reviewed the suggested changes in plan mode, and proceeded with them. Done, very straightforward.

A quick note regarding the “Documentation MCP server”: in Cursor, I feel it’s somewhat redundant given the features already built in (though please correct me if I’m wrong). Cursor supports documentation indexing and URL referencing out of the box. You can index an entire documentation site directly within the editor, and URL referencing allows you to simply paste a docs page link into the prompt window, so the agent doesn’t need to search. It just reads exactly what you provide.

It’s quite interesting to see how different AI code editors can lead you to take completely different approaches to the same task.

Cursor is better at identifying context within a project than Copilot, so there's less a developer has to do to set up that context. At the same time, I didn't have to create a separate prompt with Copilot. I could have done it all within the chat and treated it as a 1-off task.

But, consider the following:

• With the generated prompt, I can use this later to generate instructions or content for an AGENTS.md or related context files so all future generated code understands that I want to use ILogger<T> instead of IEventLogService even if the documentation, guides, or other examples I grab for context are still using IEventLogService.

• I want to share the prompt with the community because the Kentico Community Portal isn't just an Xperience by Kentico application - it's also an education resource - so everything I do with it needs to serve that purpose as well. This means sharing a detailed example of agentic software development, including the prompts, is as important (or maybe more important) than updating the repository to use ILogger<T>.

A quick note regarding the “Documentation MCP server”: in Cursor, I feel it’s somewhat redundant given the features already built in

• Not everyone uses Cursor 😅

• Having vectorized embeddings of content outside of your editor, accessible via a web service call (MCP server) can be helpful. For example, AIRA can use them!

• Copilot also supports making fetch requests to URLs, like documentation links, so this isn't a limitation. But maybe I don't want to copy and paste links into chat! This becomes more important when using something like GitHub coding agent, where you want to delegate tasks and have it use tools to complete them. "Take these tools (including the docs MCP server) and figure it out"

Thanks, Sean, great points as always. I just want to emphasize that there isn’t a single path to the same outcome. Early on, many AI coding editors felt quite similar, but they’ve since evolved to serve different audiences. Since you work in VS Code, it’s been interesting to compare that with my experience in Cursor. I’m learning a lot from this discussion and appreciate you adding context to my comments. Hopefully, a quick comparison from my side is useful to others as well.

As a developer delivering client projects, I’m very much the target audience for your articles. In that role, I don’t usually need agents reacting to GitHub issues or MCP servers sending documentation data to other services. A more integrated, in-editor experience works better for me. That’s why I find Cursor’s built-in docs indexing and URL referencing especially convenient for day-to-day work. Different workflows are valid; what matters is choosing the setup that best fits your context.

Thanks again for sharing your workflow. I’d really appreciate it if more people joined in these kinds of discussions.

I just want to emphasize that there isn’t a single path to the same outcome.

Great point. Very true!

I recommend software developers trial multiple tools (we've taken this approach internally at Kentico) and see which one helps you adopt an agentic AI mindset and approach. But, since they continue to change and evolve, check in on tool alternatives a few times a year to see what's available now.

There are many content creators that cover these tool updates (and agentic AI development technology in general). Here are some that people might find useful.

• Fireship
• Theo.gg
• Prompt Engineer
• AI Explained
• Internet of Bugs
• IndyDevDan

I’d really appreciate it if more people joined in these kinds of discussions.

Same!