The sphere of artificial intelligence (AI) is exponentially revolutionizing the global landscape, with the software development sector being a significant part of this transformation. Are you interested in augmenting your efficiency, automating routine tasks such as documentation, tests, and elementary code reviews, or even getting an opinion on whether or not AI threatens our livelihoods? Keep reading; we’ve got something for you.
We developers are already equipped with an arsenal of sophisticated tools that aid our day-to-day activities. But one may question the necessity of integrating AI; does it make a significant difference?
AI stands out due to its incorporation of natural language models (NLMs), enabling a more human-like interaction. These models are adept at maintaining conversational context, thereby eliminating the need to repetitively rephrase the same query, as often experienced when using resources such as Google or Stack Overflow directly. This results in a more seamless and rapid interaction. Another day-to-day practical advantage is that NLMs, with their anticipatory capacity, can predict your thoughts, thus offering potentially useful solutions and ideas, even without explicit solicitation. When asked about NLMs, this is what ChatGPT-4 had to say about it:
Developers can integrate my capabilities into their software to add natural language understanding and generation functionalities. This can include chatbots, virtual assistants, or any application that involves interacting with users through text or voice.
-(ChatGPT, personal communication, July 11, 2023)
Due to the AI tools’ ability to comprehend the context of our dialogue, they are capable of executing tasks that were previously challenging, or perhaps impossible, with conventional tools:
- Identification of bugs in a code snippet
- Summarization and elucidation of a piece of code
- Generation of unit tests for minor functions
- Analysis and suggestion of slight modifications to improve the code
General-Purpose AI Tools
In this section, we are going to take a look at some general-purpose tools. As the name suggests, these are not made for any specific purpose, but they can help you do most of the things previously mentioned, at least at a basic level. Here are the three most common and popular ones:
- ChatGPT is OpenAI’s chatbot that uses the GPT-4 language model, with GPT-5 estimated to be released next year. This is the most popular and perhaps most used AI model at the moment. ChatGPT generates fluent and coherent text, but it can’t access the web or other external sources (yet). It relies on a fixed dataset that was curated and pre-processed by OpenAI.
- Bard, developed by Google, is an advanced chatbot powered by LaMDA. It uses an extensive compilation of text and code, such as books, articles, code snippets, and scripts. It also leverages the endless resources of Google Search to access real-world information. It won’t be long before we see this chatbot in the medical field, as Google’s medical chatbot is already being tested in hospitals.
- Bing is Microsoft’s chatbot, which uses an upgraded version of OpenAI’s GPT-4, just like ChatGPT. It can generate text and images, search the web, and even give you sources for its answers. It has a limit of 15 turns per convo, so keep that in mind. It will soon include a “no search” feature, making responses faster and less polluted with the web’s plethora of information.
AI Tools for Devs
In the previous section, we discussed the three most popular AI chatbots made for general purposes. But the following tools are designed for the sole purpose of helping developers code faster and more efficiently:
Code completion
- GitHub Copilot A favorite one by devs everywhere. Developed by Github and OpenAI, it turns natural language prompts into coding suggestions.
- Tabnine This solution combines the power of a sophisticated public code model and a customized algorithm. This code solution learns code and patterns from your team, so it delivers business-level solutions.
- Amazon CodeWhisperer Based on thousands of millions of code lines, this one suggests code snippets to complete functions based on the existing code and on your comments.
Code review
- DeepCode is an AI-powered code review tool that stands out for its hybrid approach, combining the generative and symbolic AI with the expertise of security researchers in order to prevent hallucinations (more on this later!).
- Veracode is trained by machine learning and based on the research of security experts.
Automated Testing & QA
- Katalon Rated as one of the best AI automation testing software by Gartner, as of July 2023. You can run automated tests for mobile, desktop, APIs.
- Testim This one leverages ML for authoring, execution, and maintenance of automated test cases. It’s particularly useful for UI, as it recognizes sequences for well-architectured tests.
- LambdaTest Enjoys a great reputation among those looking for a scalable test execution platform.
There are already many tools available targeted specifically for software developers and specific roles in tech teams. If you want more information about them, you can check their websites. Remember that new AI tools are being released continuously, as well as technology improvements and new features. Keeping up with novel technologies can be overwhelming but it’s crucial!
Putting These Tools to the Test
Let’s talk about numbers and the real question behind the reason for using any of these tools: How much does my productivity increase?
GitHub released its AI-based coding assistant, Copilot, roughly 2 years ago and has made interesting discoveries since with developers who have been using it for some time. This is what their findings report:
- 88% feel more productive
- 59% feel less frustrated with coding
- 60% feel more fulfilled with their jobs
- 74% can focus on the more satisfying aspect of the job
- 88% complete tasks faster
- 96% are faster with repetitive tasks
- 77% spend less time searching or researching code
- 87% put less mental effort into repetitive tasks
This has just begun. Let’s step away from the development sector and look at a interesting case. A recent study conducted by the National Bureau of Economic Research stated that customer support agents increased their productivity by 14% by using AI tools to assist their tasks, helping dramatically those less experienced ones and helping minimally those with expertise. This parallelism is a good reference explaining AI’s impact on other fields like software development. It is only fair to assess that AI skills are essential for developers (and virtually, anyone) to stay competitive and innovative in the digital economy. They help level the playing field for less experienced programmers, older programmers, and those transitioning into new technologies, methodologies, and programming languages.
The Imperfection of AI
While the attributes mentioned above paint a utopian picture of AI as a divine entity, it is necessary to dispel such illusions and shed light on AI’s limitations.
- Accuracy. Contrary to popular belief, AI tools do not always deliver pinpoint accuracy, and errors are not infrequent. It’s advisable to cross-verify the information gleaned from AI tools. Bard, for instance, often struggles to provide verifiable sources to substantiate its claims. Traditional machines will express their incapacity to execute a task due to poor context or other reasons, AI typically will not. Instead, it tends to produce incorrect outputs. This stresses the importance of expertise in discerning the accuracy of AI-generated results.
- Artificial hallucinations. You can enhance the output of an AI tool by providing more context. Remember these tools work with LLMs that try to predict your intent and input, which can sometimes lead to divergent interpretations. This behavior is known as “hallucination,” and can manifest even in instances of abundant context. This happens when an AI confidently produces responses unsupported by its training data.
AI hallucinations may be triggered due to various factors, including these:
- Bias in training data. Hallucinations can be derived from the biased training data.
- Insufficient training data. Poor data can hinder the model’s ability to tell fact from fictitious data, thereby inducing hallucinations.
- Model complexity. As AI models grow more intricate, they become increasingly prone to hallucinations by identifying non-existent patterns in data.
- Overfitting. This is when the model is trained to effectively memorize inputs and outputs for the training data but will not be accurate for new data.
- LLMs’ vector encoding. This happens when words with more than one meaning (for instance, “bank”), have only one encoding, when they should have one per meaning. Issues with encoding and decoding could lead to hallucinations.
Is AI a Threat to Our Jobs?
The short answer is no. And there are a few reasons why we can affirm that categorically.
Firstly, an obvious disadvantage for AI tools is the lack of context about a specific organization, what we call experience in your job, for humans. While they will reduce the need for some developers, since they enable them to do more with fewer people, there is still the need for human supervision. AI is not equipped to replace software development’s creative and strategic decision-making aspects, crucial for creating innovative and user-centric applications.
The conversation is open, and it’s time to discuss how the development industry will change and what new roles are emerging because of AI advancements. Changes will come, and those ready to adapt will thrive; those who aren’t … well, might be forced to learn or change careers.
In conclusion, AI tools have undeniable advantages for developers, enabling us to do our jobs faster and better, with fewer errors, and giving us more time to be human. Be ready to surf the wave or be smashed by it. Nobody can claim there wasn’t any warning. Thus, our jobs are safe… as long as we keep up with the tools within reach and learn how to leverage them!
If you enjoyed this, be sure to check out our other AI articles.
