How has AI changed your approach to development?

August 2023 check-in — what is different in your approach in the ChatGPT, Copilot, etc. era of development?

If not much has changed, speak to that.

Comments

[Aiden Oliver]

AI has dramatically impacted the way I troubleshoot and transform API responses in the Integration space. Although it is far from being able to develop scalable and efficient technical solutions on it's own, it assists me in minor tasks such as script development and enhancements.

[Berolich]

It has really changed a lot most time when I get lost I spend hours searching the web but now it’s different 🥹

[Jon Randy 🎖️]

Not much really. I do use it, but only TabNine - which makes excellent AI suggestions based on your own code, and is like a mind reader a lot of the time).

I wouldn't touch ChatGPT or similar with a bargepole for writing code because:

ChatGPT's odds of getting code questions correct are worse than a coin flip.

ChatGPT gets code questions wrong 52% of the time • The Register

theregister.com

But I didn't need an article/study to tell me that - it became extremely obvious early on through playing with it to write code for me.

It's super dangerous using AI generators for code unless you're already a good developer - as the mistakes it makes (which are frequent) are often very hard to spot. Beginners should steer very well clear of using these tools to help them learn.

[Edenn Touitou]

It's been 4 months since I started using github copilot. I use it mainly to develop a personal design system in sveltekit (html, css, typescript).

It helped me a lot when doing repetitive tasks. Like for example, I started writing css variables like :

:root{
    --primary: rgb(59 130 246);
    --primary-dark: rgb(37 99 235);
}

And the prompt automatically completed for me the secondary, success, danger, warning and info.

When I started doing a btn class :

.btn.btn-primary {
    background-color: var(--primary-bg-button, var(--primary));
    color: var(--primary-text-button, var(--white));
    border: 1px solid var(--primary-border-button, var(--primary));
}

It took the template and used it for all the other variables.

I also used it for a small symfony (PHP) project. I started writing the controllers. When I moved on to the tests, copilot wrote lots of basic tests (some with dataProvider). I only had to tweak them a bit or to add edge cases tests.

In the end, I find it very useful for redundant tasks. And as an autocomplete on steroids. I don't think I used it in other use cases.

I don't use chatGPT, and I use very lighlty Google Bard to sometime get a help writing documentation or when I don't want to Lorem Ipsum my interfaces.

[Randall]

I feel GitHub Copilot accelerates me a lot and it does change how I go about coding. I will often pause and wait for it to auto-complete something, where I wouldn't have before. Sometimes it generates code that isn't quite what I would write, but as long as it's good code I will leave it alone. I never accept bad code from it, or code I don't understand.

I don't use ChatGPT very often, but it's in my toolbox now as a method of last resort. If I look at documentation and use my noggin and I still can't figure something out, I'll ask ChatGPT and sometimes as if by magic it understands my question and knows a great solution. So it's just one more resource for me, but it can be a pretty powerful one sometimes.

[5okol]

Despite being an experienced developer, my life has been affected. I use both ChatGPT and Copilot. In some cases ChatGPT helps a lot, for example, when integrating third-party services. You can read less documentation, just feed ChatGPT documentation and write where you want to integrate, for example NextJs.

Capailot impressed me a lot in the beginning, now it doesn't provide many solutions, mostly it helps to add some trivial things like imports.

[Ujjwal Raj]

To be honest, it really saves a lot of time for me. The only thing is that I have to keep in mind that the code writing techniques is aligned with the project I am contributing to or building. I basically look into the crux idea how AI solves the issue I want to solve and get idea from their code. Other than that I have to write my own code. But anyways, this only saves a lot of time for me. At least 3 times faster development.

[Maggie Martin]

I've started out mostly using it for refactoring smaller chunks of code - first explaining what I want to do, then pasting my current code. From there, I test it and review it to see if it actually does what I want, and if it is, in fact, a better written/more condensed version of my original. I may ask them to rewrite it and see some additional options. I probably use their suggestions about 1/3rd of the time.

[Marcin Rybacki]

Copilot has been a bit disappointing for me. It turns out I don't do so many repetitive tasks that this tool can relieve me from...

[Peter Harrison]

Have included it in my workflow for some time. Useful so long as you understand its limits and respect that it lies like a weasel when it reaches the limit of its abilities. Always double check generated code, and learn from it rather than cut and paste. May not be replacing developers yet, but is pretty impressive.