Unity + AI coding tools - current state? (June 2025)

Hi everyone,

I’m curious about the current state (June 2025) of AI integration with Unity - specifically for code generation and assistance within the Unity ecosystem.

I’m aware that in web development, tools like Claude can already analyze your entire codebase, understand project context and structure and allow you to generate or modify code through prompts with fairly good awareness of your architecture.

I’m wondering: Is anything comparable available or usable within Unity?

Ideally, I’d like a workflow where I can:

  • Import an AI assistant into my Unity project
  • Let it analyze the entire codebase (in my case, a large project written using Unity ECS/DOTS)
  • Prompt it to write or edit code based on that understanding

Are there any current tools, plugins, or workflows (maybe even experimental ones) that can do this reasonably well in Unity? Has anyone tried integrating tools like Claude, Copilot, or ChatGPT in such a manner and what are your results?

I’d love to hear your experiences or recommendations, especially from folks working with complex or data-oriented projects.

Thanks in advance!

3 Likes

I recently published this video that compares a few AI assistants for Unity:
Unity AI Comparison - Unity vs. JetBrains vs. Bezi
Hope that is helpful!

3 Likes

As far as I have researched, haven’t seen any as mature as web dev in Unity, so I am personally exploring the development of the Agentic Assistant Plugin.

  • Agentic Assistant Plugin
    • It’s in early beta, which generates UI natively inside the editor. It doesn’t analyze whole projects or DOTS yet, but it’s useful for rapidly prototyping or iterating UI.
  • Unity awareness through RAG
    • This one utilizes RAG to provide Unity contexts to the LLM, which will be embedded into the above plugin in near future for context-aware AI assistance.

Do you think they can help solve your problems?

1 Like

You should take these AI tools with a grain (no, a sack) of salt. And then maybe just throwing Claude Code into the mix of whatever AI flavor you’re using. They just seem to always be hit or miss for me

1 Like

I have a big DOTS projects as well. I use AI but only to generate methods or something like “prepare and schedule this job”, then they fill in the required parameters. They are an annoyance when they start touching multiple parts in a file, or worse, files. The code is almost always wrong. I think they don’t have much DOTS training data and they suck in this domain. I’ve read from others that AI code generation is generally not effective in big project or projects in maintenance. You add a niche framework to that and what do you get?

AI can slow experienced developers by ~19% in familiar code, but boost beginners’ productivity by 20–30% by providing guidance and reducing trial-and-error.

The question is whether relying on AI over the long term might weaken our memory, similar to what happened with Google search, where we often remember the phrases we used to find something rather than the actual information we were looking for.

I’m using Cursor Agents (sonnet-4 \ gpt5) with Unity codebase lately on a daily basis.

  1. It still sucks at UI Toolkit authoring - confuses USS with CSS, can’t build components and complex UI layouts in mxml without issues even if I add latest Ui Toolkit documentation as the context.

  2. It helps make large-scale changes to the codebase at both DOTS and non-DOTS contexts.

  3. Helps spot bottlenecks, understand a new codebase architecture (can draw architecture overview etc). Can sometimes miss assembly references when generating asmdefs, but overall helpful, especially when landing to the new tech / stack either within Unity or any other tech.

  4. You can augment your AI-powered IDE with an MCP server, such as this MIT-licensed one allowing more automation and some basic actions performed by AI agent within your unity project.

1 Like

Yet it’s senior developers who use AI more: Senior developers let AI do more of the coding — survey | InfoWorld
A lot of contradicting data out there. It appears that it’s too early to draw firm conclusions.

From personal experience, AI simply with copilot, made development more pleasant. ChatGPT for self-contained blocks like a particular algorithm or shader.

2 Likes

I was going to write my own experience but it is basically the same!
Copilot has improved auto completion but can hallucinate and suggest non-existent functions, ChatGPT is a freely available but error-prone source for general code creation and more.

I have not used it much to review code, but it seems copilot can spot many issues.

But this about plugins for your coding IDE, not specifically Unity - there is another space for in-editor Unity work which Unity is trying to fill itself but I need that less.