Over the past year, AI coding assistants like GitHub Copilot, ChatGPT, Cursor, and Claude have become part of many developers’ workflows. Some teams report significant productivity gains, while others argue that AI mainly speeds up boilerplate code but still requires careful review.
I’m curious about real-world experiences.
Has AI genuinely made you more productive?
Which tasks do you trust AI with (coding, debugging, documentation, testing, code reviews, etc.)?
Have you noticed any downsides, such as incorrect code, security concerns, or over-reliance?
Do you think AI will change the role of software developers over the next few years, or will it simply become another tool in the development stack?
I’d love to hear how you’re using AI in your day-to-day development work and whether the benefits outweigh the challenges.
Yes, both regarding coding speed and thinking outside the box. I have tested Vibe coding, but found that being a mess. Fast, but almost impossible to maintain or understand.
All but documentation.
If I write good prompts, I get brilliant code sometimes. Sometimes a complete mess. And we do have different perspective on security. AI is addictive.
How knows? Vibe coding is conveniently, but when I asked AI itself about Vibe Coding I got some warnings about short term thinking (what happens 5+ years from now with your app?)
AI is an electric wheelchair sort of. Comfortable. But used too much it may resulting in that you cannot walk by yourself in the future…
Definitely not documentation, testing* or code reviews*
*: If the originator is AI. Never remove a human from the cycle entirely.
In my own code: Logical errors.
In others code: Over-reliance is a MASSIVE problem.
Those two things are not mutually exclusive. That said, I do think (like pretty much any new technology) we’re currently riding in a bubble, that will deflate. However, it wont deflate back to pre-AI, it will find its ‘normal’ place.
I’m curious. Are you counting the amount of time it takes you to craft the prompt in the coding speed? Because that’s a negative towards your productivity. You are “coding” your prompt sufficiently verbosely for the AI to generate what you want.
You must be a great writer then, if you’re able to get a full fledged prompt for an AI to write you some brand new code in less than a minute and actually have it come up with something that matches what you wanted.
In my opinion yes, it increase productivity if use properly; but don’t believe the hype that vendors of AI tell you of increases of 10x; they are trying hard for you to buy their service.
For example let said AI write the code, you have to carefully review it, so you become the bottleneck, so to speak.
Another thing that as @Sibertius mention, if you overrelied on AI, your skills atrophic. The have been numerous studies to support that.
And remember when the 2:00 am call come, they are not going to call AI, there are going to call you.
Yes coding at least. I do not know about the others except virtual assistants have been very helpful in improving specifications I have developed prior to getting the program generated.
Definitely, in many ways. I am not aware of generated code having bugs during execution but virtual assistants do sometimes hallucinate, such as hallucinating methods that do not exist, that can make the entire program useless since the library it uses is useless for the requirements.
There will be changes. Developers comfortable with AI will likely be more successful. It is extremely tempting to just use generated code without thoroughly analyzing it but there are multiple reasons why we should analyze the code. We need to understand the code and if it uses features we are unfamiliar with then we need to learn. We need to ensure it does what we intended it to do.
I think another way that virtual assistants are helpful is to compensate for poor documentation. Much of the documentation I see these days are not clear. Software people have gotten lazy about documentation. Documentation of a ReadChunk method might just say reads a chunk of data. That problem should not exist but it does. For simple methods such as that it is not much of a problem but more complex methods and classes might not be understandable. A virtual assistant can save much time by showing use of software that might not be documented well.
