GitHub Copilot Struggles
AIM Research
Strategic insights for Artificial Intelligence Industry. For Brand collaborations, write to [email protected]
An ML teacher and founder of Tideily, Santiago Valdarrama , recently asked the question – “If you aren’t using GitHub Copilot to write code, can you tell me why?” Surprisingly, the responses were mixed, and the insights were nothing short of fascinating.
Copilot’s bitter sweet symphony?
Tom Cruickshank , a developer who tested Copilot with JetBrains IDE, shared his experience. “I tried it out with Jetbrains IDE and I felt it was more of a miss than a hit. I was deleting stuff more than using what it was giving. And I could not justify the price for the value I was getting. Tried it for a month. Removed it, Cancelled subscription.”?
Cruickshank’s frustration highlights a common sentiment among developers who feel that the tool’s output often misses the mark, leading to more work rather than less. Developers often spend more time debugging the code than actually using it within production.?
Another primary complaint from experienced developers is that Copilot disrupts their workflow. Darren Shepherd , an accomplished coder, pointed out, “For languages I am very experienced in, it interrupts my mental flow with long suggestions. I don’t really need help on what to do, I know what I’m doing, I need help on names/references that I don’t remember.?
Serial entrepreneur Danny Lieberman believes that one can code a lot faster if they know the patterns, anti-design patterns and solutions. “You can also get a lot of mileage from Warp without copilot,” he added, saying that he is sort of like ‘coding warm turkey,’ – “I don’t use copilot but I find ChatGPT useful for reminding me how to do things.”?
Coders’ nightmare?
While Copilot might be helpful for simpler tasks, many developers find it lacking in more complex scenarios. Mark, another developer, commented, “LLMs are great for a particular type of coding, not all, by far.”
The use of Copilot also raises concerns about the learning process for new developers. Ross Murphy expressed a cautionary note: “If you ever want to create anything new, you’re going to need to understand how to actually write code, and not just use a glorified copy-paste tool.”
Using GitHub Copilot is one sure-fire way to never actually learn how to do coding. “Because I have no clue what I am doing. I don’t need autocomplete - I need to tell Claude ‘This is what I wanna do’ and it creates the code for me.? I’m learning a lot – but I don’t have a baseline of knowledge,” mentioned a user on X.?
领英推荐
Developers emphasise the importance of maintaining a clear mental model of their code. Copilot is not very useful for anything beyond auto complete. As a programmer, you want to have full context in your head of how a chunk of code works. Sometimes, Copilot suggests something more advanced and ends up in the waste of several hours.?
Similarly, AI tools like Copilot are known to sometimes generate incorrect or nonsensical code, a phenomenon known as “hallucination.” Han MF Brolo simply stated, “It hallucinates too much.” This issue can be particularly frustrating, as it forces developers to spend additional time reviewing and correcting code that should have been accurate from the start.
Additionally, since Copilot is trained on public code, it might generate snippets that resemble copyrighted or licensed code, potentially causing legal issues if used inappropriately, particularly in commercial projects.
Lauds Claude and ChatGPT
Many developers said that GitHub Copilot may also generate code that doesn’t match a team’s coding standards or architecture, leading to inconsistencies and requiring extra time for refactoring and review. People have been frustrated with this and have slowly stopped adopting such AI tools.
Given these challenges, many developers are turning to alternatives or choosing to code without AI assistance altogether. Some have found more satisfaction with other tools: “I use Cursor. I’ve tested both, and I must admit that Cursor is much better. Since I started using it, I’ve been more productive, I manage to do a lot more in a workday. It’s impressive.”
This suggests that while Copilot may not be the perfect fit for everyone, other tools might better align with certain workflows and preferences. Some people simply want chatbots like ChatGPT or Claude to generate the whole code as they don’t know what they are doing in the first place.
Enjoy the full story here .?