AutoDev Next: IDE as AI programming service, building multi-terminal programming

The core concept of AutoDev Next is to combine powerful IDE capabilities with AI to create a next-generation AI programming experience of "IDE as a service" and a programming service experience for creating applications anywhere.

GitHub : https://github.com/unit-mesh/auto-dev-next

Introduction: The Continuous Evolution of AI Programming

We are experiencing a transition from traditional manual coding to AI-assisted development. In this new paradigm, developers can describe their needs in natural language, and AI will be responsible for generating the corresponding code. The role of developers is gradually changing from a simple code writer to an AI instructor, code reviewer, and ultimate improver.

Vibe Coding: A new paradigm for programming driven by natural language

"Vibe Coding" is an important programming technology in this trend. Its core is to use natural language prompts to guide AI to generate code, and continuously improve it through a conversational and iterative approach ( description- > generation- > review- > feedback- > modification ). This method aims to lower the threshold of programming and allow developers to focus more on "what to do" rather than "how to write accurately", reflecting the trend of "natural language becoming a popular programming language". Of course, due to the lack of rigor in AI code generation, it is easy to lead to " shit- fence programming "

The evolution and challenges of AI programming tools

In the past few months, we have investigated various mainstream AI programming tools, which can be roughly divided into:

• Pan-programming : Allow non-professionals to generate applications through natural language (such as Lovable, V0).

• Professional Programming : Assist professional developers to generate code snippets through chat (such as Cursor, Copilot).

• Innovative Exploration Programming : Explore other AI programming possibilities (such as Claude Code, MetaGPT).

Although these tools have their own advantages, they are often limited to specific scenarios or interaction modes, fail to fully utilize deep project context, and have difficulty covering developers' end-to-end complex needs, which leaves room for providing a deeper and smarter Vibe Coding experience.

Trend: Exploration direction of AutoDev 2.0

Based on our observations of existing tools and our thoughts on the future of AI programming, we conducted a proof-of-concept (POC) exploration of AutoDev Next, aiming to deepen the Vibe Coding experience through the “IDE as a Service” model:

1. AutoDev Planner: Mastering Complex Programming Tasks

Although large models are good at generating code snippets, they require stronger planning capabilities when faced with complex tasks. AutoDev Planner uses advanced reasoning models (such as DeepSeek R1) to  generate a detailed coding task plan after receiving the user's natural language requirements, combined with the deep project context provided by the IDE . This plan is visible and dynamic , the AI ​​can be adjusted during execution, and the user can clearly see the progress and even manually intervene or execute steps . This brings structure and controllability to Vibe Coding, going beyond simple code generation and extending AI capabilities to the task planning and execution level, enabling it to better cope with large or ambiguous development needs.

2. MCP + IDE as a Service: Expanding the boundaries of programming intelligence

In order to enable AI to better understand and utilize the numerous tools and services in modern development, we embraced MCP (Model Context Protocol). By implementing AutoDev Next as an MCP server, it can provide other Agent tools with high-quality context information inside the IDE (such as code structure, symbol definition, debugging information, etc.). This means that even Agent-centric automation processes can call AutoDev Next to complete imperative operations that require sophisticated IDE understanding , seamlessly integrating the capabilities of AI agents into a wider range of automated workflows.

AutoDev Next: Cross-platform programming service

As the next generation AI programming tool, AutoDev Next's core feature and goal is to provide a comprehensive cross-platform IDE-as-a-service solution. It aims to significantly improve development efficiency, code quality and development experience by deeply integrating AI technology into the development process.

IDE as a backend cloud service: based on Intellij Community open source version

Its core is "IDE as a service", which means weakening the local presence of IDE and using it as a service. In this way, AutoDev Next can be used as an asynchronous programming server, automatically generating code related to your requirements when you start a requirement, and providing context support for your local code; it can also be used as a cloud service to be called by other Agent tools.

At the same time, combined with the capabilities of Planner, you can view the current task progress on your phone or in other Agent tools. When a task fails, you can remind the model of some new context information to help the model better understand the current task.

IntelliJ Community was chosen as the backend service for AutoDev Next based on its powerful features and flexibility as an open source IDE.

Multi-terminal support: Compose UI creates cross-platform programming services

In order to provide a modern, efficient and cross-platform consistent  Vibe Coding interactive interface , the UI of AutoDev Next is refactored using JetBrains' Compose UI (Compose Multiplatform). Compared with the traditional WebView solution, this Kotlin-based declarative UI framework can provide better performance, more native experience and more flexible design capabilities. Its goal is to achieve "write once, run on multiple terminals", covering Android, iOS, Web, IntelliJ IDEA plug-in itself and desktop applications, so that developers can get smooth and consistent Vibe Coding conversations and code review experiences on different devices . Therefore, you will have the opportunity to see the birth of new versions of AutoDev Android and AutoDev Web.

Preventing Crappy Programming: AI Code Quality Guardian

Due to the lack of rigor in AI code generation, it is easy to cause "shit-fence programming". To this end, AutoDev Next has designed an automated mechanism to ensure code quality:

1. AI Code Review (AutoDev Guardian) : During the Vibe Coding process, AI automatically performs code reviews, detects logic errors, code style, and potential technical debt, and provides improvement suggestions.

2. Intelligent code scoring : Combined with LLM to evaluate code quality, it provides A/B/C/D ratings to help developers intuitively determine whether AI code is usable.

3. Context-aware comparison : Analyze the differences between new and old codes to detect whether redundancy, unreasonable design or potential bugs are introduced.

4. Smart Debug & Rollback : Combined with IDE debugging tools, it automatically detects exceptions, recommends repair solutions, and supports one-click rollback of optimized versions.

5. AutoDev Quality Rule Library : Provides customizable code quality checking rules to adapt to the development standards of different teams.

These features ensure that AI code is both efficient and maintainable, truly achieving intelligent, high-quality code.

total