AI Code Assistants Reviews and Ratings

The introduction of GitHub Copilot to our internal teams was very smooth, thanks in large part to the consistent support provided by GitHub throughout the transition. The product was quickly adopted by our staff, making it one of the more popular tools we've rolled out. User satisfaction has been notably positive, and I'm pleased to see the product continually improving. Our teams have embraced the code suggestions from GitHub Copilot. We have close to 3 million lines of recommended code accepted. As we approach the one-year mark since our initial launch, we've benefitted from GitHub's commitment to our success, including comprehensive training sessions for our teams. These sessions have ranged from general introductions to Copilot to more specialized training for specific teams, and advanced prompt engineering. The training provided by GitHub has played a crucial role in enhancing our team's proficiency with GitHub Copilot. We're appreciative of the support and education that have helped us make the most of this tool, and we look forward to further developments as we continue to use Copilot in our daily work. Please note at the moment we are using GitHub Copilot Business, but we intend to upgrade to GitHub Copilot Enterprise next month. Some of the features below are unlocked with that edition but I cannot comment as we are not using those features yet.

We decided to try out Cody and immediately noticed significant improvements. Cody from Sourcegraph demonstrates a superior understanding of entirely larger codebases and provides more precise and technically more accurate answers to our queries. One standout feature of Cody is the ability to choose between popular large language models (LLMs), allowing us to research and select the best model for our specific needs. Additionally, Cody enables us to save and reuse commands for recurring tasks, which enhances our efficiency and workflow. Cody is always up-to-date and delivering rapid updates as new and better LLMs for coding are established in the fast-evolving AI field. This ensures that we are always working with the latest and most effective tools available to deliver the best coding assistance to our developers.

I have used Gemini Code assist (deut ai) for almost a year mainly to get help with google cloud services i.e. Design solutions, provision code snippets and cloud issue troubleshooting with high rates of success. I am also using this tool for software application development with average correctness.

Support is good. They answer quite fast. It's a very open-minded company and great that nearly all development is open sourced. Code Suggestions are not as good as expected.

We equipped our development teams with Codeium and realised decent impact on productivity. However, the amount of impact varied from project and function, and the average came out to 20-30% in application development with programming heavy tasks. Auto code completion functionality and snippet generation proved to be immensely useful while developing larger functions, developers regularly test smaller snippets in silos and they we're quickly auto-generated. Another area where efficiency was realised is writing unit test cases.

I used Tabnine as an extension in Code for about 4 months. The overall experience was not good due to the user interface and how it interacts with the code. While it didn't feel like any features were missing and it also provided a "secure" way to use an AI in private mode - which I doubt - the overall experience wasn't great.