Should you be coding with generative AI models?

What is generative AI?

Artificial intelligence is useful in many scenarios. AI uses algorithms and data to complete tasks which would usually require human thought processes. This is done by collecting data and adding algorithms to this data. AI is able to gather patterns from the data and apply it to future requests. Then, it makes predictions based solely on the data it's already analyzed. As a subcategory of AI, generative AI seems to be the best of both worlds. When it comes to GenAI, the same process applies. Data is collected and analyzed, but GenAI goes one step further. These models can also create new data based solely what they were trained on. The main purpose of generative AI is to develop text, images, and music based on how they were trained. These models take raw data and generate the most likely outcomes when asked to do so. This new data can even include code. This can be a time saver for busy web devs if the code produced is properly formulated. Although AI has become a popular tool for almost everything, how accurate is the code that GenAI produces?

Using generative AI to code

You can, in fact, use generative AI to code. However, in order to make the coding experience more time efficient, its important to understand how the AI works and which tools are best suited to meet your coding needs. Understanding the history of GenAI can also be useful to ensure you are using it to its full capability.

Although the roots of GenAI date all the way back to the 1960s, it still seems to be a relatively new phenomenon to web devs today. It wasn't until the 2020s when generative AI took off and many realized the coding potential. Generative AI models have also been used for many years by larger companies to help interpret and produce data such as statistics. These models have been widely used since before the 2010s.

There seems to be a lot of potential for GenAI to benefit coders. Let's break down how it works. When a user of GenAI enters a prompt explaining what it wants the code to do, they should receive that code from the model. But that's not all. GenAI can also suggest improvements to existing code, modernize code, translate code, and help developers learn new code. The Large Language Models (LLMs) have been trained on existing code. This is where the user will enter the prompt into. They will receive the code either in its entirety, or the user can ask just to be provided with snippets.

Let's back up a little bit and discuss LLMs. LLMs are AI programs such as ChatGPT. They are trained on large amounts of data allowing them to make interpretations and predictions. LLMs are based on machine learning, where such high volumes of data are fed into them that they are able to break down the data without the help of humans. LLMs are trained to respond to prompts by the user in a way which makes sense. Unfortunately, LLMs still make mistakes and are not 100% reliable. LLMs can also have bugs and can be manipulated like most other applications.

Types of generative AI for coding

With the many different models out there, which GenAI tools are best suited for coding? We combined our knowledge with that of multiple sources. Here's what we found.

• ChatGPT-4.0:
  • According to Olaf Thielke of Medium, it was discovered that when using ChatGPT to code, it can take multiple prompts for ChatGPT to produce somewhat sufficient code. While ChatGPT can generate basic code, its nothing special. Thielke concluded that most developers will be comfortable with the code that ChatGPT generates, however, using a human programmer is more likely to produce better results.
  • Our thoughts: we find ChatGPT to be good for small coding issues. However, this code needs to be reviewed. While ChatGPT does provide fast solutions, it often loops (if you ask it to fix one thing, it will undo the other issues it already fixed). To conclude, we would say that ChatGPT is very useful, but not a replacement for a human coder. It should be used as an assistance tool and in a way can almost replace Google. ChatGPT is much quicker than Google at finding coding answers and provides a more direct answer.
• Amazon Q Developer:
  • Martin Heller of Infoworld gathered that Amazon Q Developer is good at completing lines of code and doc strings. However, it has trouble generating full functions for certain cases. Heller also found that while Amazon Q Developer did have some mistakes, he would recommend it for experienced developers. According to Heller, Amazon Q Developer also scans for vulnerabilities which can help developers fix potential issues.
  • Our thoughts: the first major different between ChatGPT and Amazon Q Developer is that AQD is directly embedded in the IDE as an extension. This use case isn't the best for our day to day, as we're often not coding directly into an editor but instead using a web tool such as Webflow for our clients. The code generation is also slower than ChatGPT-4.0. However, it is nice to have direct access to the codebase. AQD will be a tool that we will definitely something we will keep experimenting with.
• Gemini 2.5 Pro:
  • Gemini 2.5 Pro was released at the end of March and yielded surprising results. ForrestKnight deemed Gemini 2.5 Pro as "the best coding AI I've ever used, and I've used them all." Theo.t3.gg raves about Gemini 2.5 Pro's speed of responses, performance reviews, benchmarks set, and more. Theo even stated that Gemini 2.5 Pro is "slaughtering everything else on the market right now."
  • Our thoughts: following the release of Gemini 2.5 Pro, we are still in the process of testing it out. So far we have recognized its great qualities and how much it has improved from its predecessor. One of the best parts of Gemini is that its directly embedded in our Google Workspace making it very convenient for the whole team. It has AMAZING features that ChatGPT-4.0 doesn't offer and overall, we were blown away. The new version of Gemini is one of our new favourite tools.

While we didn't test every other generative AI model out there, we did choose some of the most basic and popular ones. There was definitely a clear winner. Web devs have been raving about their experiences with the new Gemini model and we can see why. While some have stated that Gemini 2.5 Pro narrowly beats out ChatGPT-4.0, others would disagree. It's clear that Gemini 2.5 Pro has the highest stats right now and many are saying that Google has won the AI battle.

Should you code with generative AI?

While the models have been getting steadily better over time, many web devs remind us that these tools should be used as an assistant for coding and not as a replacement for human coders. GenAI models can greatly speed up the coding process and help reduce repetitiveness for humans. However, they can still make mistakes. The bottom line is that generative AI is a great tool to speed up time and allow you to focus on more difficult aspects of coding, just be sure to double check for mistakes and consider giving today's AI models easier tasks. Sometime in the future, generative AI will be able to to do all the coding for us, and that time is close.

Environmental impacts of generative AI

It's clear that GenAI has positively impacted the world in many ways and not just through coding. However, what are the environmental impacts of GenAI? According to science, its not good. Generative AI uses a lot of resources, mostly energy and water. Generative AI uses a humongous amount of energy. On top of this, the construction of data centres has increased drastically. In addition, water is used to cool the systems used for training the models and this is heavily impacting many ecosystems. Some company's water usage has increased by almost 35% as they create AI models. What's even worse is that as different versions of GenAI are being released constantly, all of that previous energy use is being wasted. As scientists work towards making GenAI a more sustainable industry, these issues are becoming increasingly worrisome.

Wrap-up

As we all know, GenAI has been taking the world by storm. Many models are more than fit to be used for coding (with some careful revision by humans of course). However, as generative AI grows, so do the environmental concerns that come along with it. Hopefully, we will get to a point in the future when GenAI becomes an industry that can do more good than harm to our planet.

As always, thank you for taking the time to read our blog today. If you have any questions or concerns, please feel free to email us at contact@gemify.ca. If you would like to see more from us, please check out our Instagram and LinkedIn pages and subscribe to our newsletter. We used many great resources for this blog, so we've listed them below for you to view.

Resources (in order of appearance)

Olaf Thielk's article on ChatGPT
Martin Heller's review on Amazon Q Developer

ForrestKnight's YouTube video on Gemini 2.5 Pro

Theo-t3.gg's YouTube video on Gemini 2.5 Pro