The GitHub Copilot Workspace beta failed to complete a simple task, performing worse than expected at every step. It didn't follow conventions, wrote information to Markdown instead of JSON, and missed basic details. The tool's workflow is convoluted, creating more work than it saves. Despite reattempts with specific instructions, the results were still unsatisfactory. The review concludes that while the tool may benefit someone who dislikes programming, it ultimately wastes time with broken, inaccurate outputs.
If the summary seems innacurate, just downvote and I'll try to delete the comment eventually 👍
Maybe you could give our newly launched Autonomous AI Programming Assistant: Clacky AI a try. It provides a complete cloud-based development environment where AI can autonomously handle the entire development workflow:
- Requirement analysis
- Plan generation
- Code execution
- Change review and modifications
- Commit creation
- Pull request generation
When it comes to creating new code, my experience with copilot has been about the same. Honestly pretty bad at even autocomplete. Noticeably worse than what we've had in the past in Visual Studio that wasn't called AI.
Only decent AI coding I could manage was locally ran Llama 3 with Ollama and sending my entire codebase and documentation as RAG. Got better results, but it's still.. not great.
I test AmazonQ at work rn and pasted a simple json and asked it to create a java record that could be parsed from it and it fucked up even this. I just don't know how those LLMs can be considered reliable.
fagnerbrack@reddit (OP)
Here's what you need to know:
The GitHub Copilot Workspace beta failed to complete a simple task, performing worse than expected at every step. It didn't follow conventions, wrote information to Markdown instead of JSON, and missed basic details. The tool's workflow is convoluted, creating more work than it saves. Despite reattempts with specific instructions, the results were still unsatisfactory. The review concludes that while the tool may benefit someone who dislikes programming, it ultimately wastes time with broken, inaccurate outputs.
If the summary seems innacurate, just downvote and I'll try to delete the comment eventually 👍
^(Click here for more info, I read all comments)
Narrow_Tradition_305@reddit
How did you get the access? I'm in the waitlist for months
Educational-Noise168@reddit
Maybe you could give our newly launched Autonomous AI Programming Assistant: Clacky AI a try. It provides a complete cloud-based development environment where AI can autonomously handle the entire development workflow: - Requirement analysis - Plan generation - Code execution - Change review and modifications - Commit creation - Pull request generation
Quick demo video: [https://youtu.be/bjngOz688YA]
fagnerbrack@reddit (OP)
I'm not the author just sharing
10lbCheeseBurger@reddit
Claude is the only model I've found to be consistently good. Copilot has been a bummer for me since it launched.
geepytee@reddit
Github Copilot disappoints again :D
Letiferr@reddit
When it comes to creating new code, my experience with copilot has been about the same. Honestly pretty bad at even autocomplete. Noticeably worse than what we've had in the past in Visual Studio that wasn't called AI.
krileon@reddit
Only decent AI coding I could manage was locally ran Llama 3 with Ollama and sending my entire codebase and documentation as RAG. Got better results, but it's still.. not great.
Incorrect_ASSertion@reddit
I test AmazonQ at work rn and pasted a simple json and asked it to create a java record that could be parsed from it and it fucked up even this. I just don't know how those LLMs can be considered reliable.