AI assisted programming is here. But as of today, AI technologies are unreliable and they lack the necessary accuracy for refactoring source code. The burden is still on developers to check the generated code to ensure it isn't buggy or adding to overall poor code quality. By themselves, AI assisted programming tools just aren't good enough.
That's why we created our new automated refactoring tool with fact checking code validation, now in beta.
More precise
Not only did we train an AI model to write code for refactoring, we trained a secondary AI to fact check it
More impactful
By automating code refactoring, dev teams can vastly improve their overall code quality while still continuing to innovate
Calling all early adopters
Now accepting sign ups to the waitlist
Join the Beta testing program for CodeScene's new AI generated code refactoring tool.
You'll get first access to our proprietary technology that is the only generative AI coding assistant with up to 97% accuracy.
Early access starts in January 2024. We will accept participants gradually.
Please read the terms and conditions:
The first languages available are JavaScript and TypeScript. If you use other languages, please indicate them in the form. It can help motivate our future roadmap.
You must have an active CodeScene license to participate in the beta test. Either by being a current customer, or by signing up for a free trial at the start of the testing phase.
By signing up to the waitlist you are agreeing to participate in feedback surveys, and potentially to be selected for focus groups and interviews for internal use and/or marketing.
Backed by research
With a benchmark study, CodeScene's Adam Tornhill, Markus Borg and Enys Mones explore a new frontier by investigating AI support for improving existing code.
The majority of a developer’s time isn't writing but understanding and maintaining existing code. But today’s AI is simply too error-prone, and far from a point where it is able to securely modify existing code.
In this whitepaper, they benchmark the performance of the most popular Large-Language Models (LLM) on refactoring tasks for improving real-world code, with shocking results. Read the study and learn how CodeScene uses a different approach to ensure accuracy.
Relevance, context and impact
How does it work?
Understand the health of your entire code base
We use our proprietary algorithm called Code Health to identify refactoring targets in your code base based on a number of contextual factors and how your code has evolved over time. It is the only code-level metric with proven business impact
Get refactoring suggestions tailored to code smell
Not only does CodeScene identify what code is low in quality or has less than perfect Code Health, we understand why, by identifying a number of the most common code smells
Use generative AI to refactor legacy code
Using this gold standard metric, the AI then produces a recommended replacement for the code snippet that should be improved, ensuring the AI written code does, in fact, fix the code smells
Accurate and robust training model
The model has been trained on a massive data lake of refactoring examples so that it recognizes what old "bad" code looks like, as well as the changes made to improve it
Fact check AI generated code
We then use a secondary CodeScene AI that fact checks the AI generated refactored code. In early tests, we see a 97% accuracy rate, a stark difference to the current industry benchmark of 40-50%
Same certified security
We never store your source code, nor do we use it to train our models. We also never send the full source code to the AI, only snippets. Everything is encrypted and protected in transit
Learn more about CodeScene's Code Health and refactoring targets
Our AI for automated refactoring is built on top of our already leading software for code analysis and technical debt reduction.
