Cursor’s new visual editor: Right idea, wrong implementation

Andreas Møller

December 11, 2025

Read the release announcement.

Why add a visual editor to an AI coding tool?

LLMs are incredible at generating large amounts of code faster than any programmer can type. For a select number of tasks they far outperform humans. They are especially useful when two criteria are met:

1. When the problem to be solved follows a similar pattern to ones the LLM has seen in its training data.
2. When the desired outcome can be clearly defined. For example, “Write a function that finds the first element in the basket where the price including VAT is larger than €45.”

Criteria number 1 is purely up to the model, so there is not much that Cursor or any other AI-dependent tool can do about that. As models get better LLMs will be able to handle more cases, but this limitation will likely always exist to some extent.

Criteria number 2 is what Cursor is attempting to address with this new update. People who have been coding with LLMs for a while have probably seen that they tend to do well at back-end tasks but less so on the front end. When creating user interfaces, the first prompt often gets you 80% of the way to what you want, but the remaining 20% can be painful. You often have to get increasingly detailed about what exactly the LLM should do, and you often have to go through several prompt cycles getting more and more specific in your request.

Where back-end tasks often have only a few valid solutions, front-end and user interface programming is different. There are often dozens, if not hundreds of ways to accomplish the same thing, and it is hard to objectively say which of them are “correct” and which are not.

Ask any flagship model to: “Write me a function that takes a string as input and copies it to the user’s clipboard” and they will get it right 100% of the time. Ask them to create the button that actually copies that string and they are unlikely to produce the exact style, size and icon you were looking for.

The problem is not the LLM: it is that your request was not precise. You're prompting it wrong, bro! There were simply too many possible valid solutions for it to pick the one you were thinking of. When the LLMs can’t predict what you want, you have to tell it explicitly, and this is where the visual editor comes in.

The new visual editor in Cursor Browser allows you to make visual changes to your interface that are then converted into a prompt that precisely tells the LLM what changes to make. Making the same kind of detailed changes by prompting would require several iterations, waiting for the LLM at every step. The visual editor makes this process both easier and much faster.

AI and visual development: the perfect match

The problem Cursor is trying to solve is likely something you have experienced yourself.

While LLMs are excellent for the bulk of code generation, especially when it comes to commonly understood problems and repetitive tasks, we still need tools that allow us to do the fine-detailed work that makes our websites and apps special. If an LLM is a chainsaw, then visual editors are the carefully sharpened chisels. The key is learning when to use which.

We have previously written about the many benefits of visual editors when creating user interfaces. You can read more here.

While Cursor's solution is clever it does pose an obvious paradox:

All of the tokens, none of the benefits

Cursor identified the right problem to solve. But their solution has none of the benefits we normally get from using LLMs, while still having all of the drawbacks. You still need to give Cursor the exact same precise instructions as if you were writing code, but by running your set of changes through an LLM, you not only have to wait for the LLM to apply those changes, you also have to deal with the indeterminism that AI models inherently impose. By running your precise set of instructions through LLMs you open them up to interpretation, which means you still have to verify that the resulting code is exactly what you intended.

As if all of these issues were not enough on their own, there is one more ironical cherry on top. Even though you did all the work, you still have to pay for the tokens. The promise of AI was that it would do the work for us. Instead, we are having to pay for work we did ourselves.

Nordcraft is designed from first principles

Cursor's vision is not that far from the one we have at Nordcraft. The next big innovation in web development is going to be focused on how we write code. Combining the raw power of LLMs with the precision of a visual editor creates unique opportunities and can remove a lot of the pain points of traditional coding.

Cursor's approach comes from incremental thinking: each new feature trying to fix the problems they created with the previous. At Nordcraft we took a step back and redesigned the entire stack. Every part of Nordcraft is designed for this new way of working.

The Nordcraft AI assistant will generate full pages in a manner of seconds, and the visual editor will let you make detailed changes fast and efficiently, without using extra tokens. The web framework that powers Nordcraft applications and the editor are designed for each other, so each change you make is committed instantly without any chance for an LLM to misinterpret your intentions and without wasting precious unnecessary tokens.

Nordcraft’s AI Assistant does not just generate large swathes of code for you, it will carefully take you through how every part works. And, Nordcraft's visual coding interface makes it simple to understand the logic and behavior of your application so that you are always in control of what you publish. 

LLMs are a fantastic tool that can propel you forward faster than was possible just a few years ago, but the things that really matter, the things that make your product stand out from every other product out there, have to come from you.

Nordcraft AI is coming early 2026. Sign up now to get notified!