Whitepaper – A programmer’s guide to AI (Part 1)

Software development is constantly evolving, and the past few decades have seen major shifts in the way we build digital products. This evolution enables us to develop software faster and better.

Most well-known evolutions in software development are the replacement of waterfall to an agile way of organizing our work and the introduction of continuous integration and delivery (CI/CD). The essence of these evolutions is to shorten the feedback cycle, both in development with Agile methodologies and in delivery with CI/CD techniques.

GitHub’s research shows that AI Co-Piloting can significantly increase developer productivity, allowing them to complete tasks up to 55% faster. This remarkable speed increase is attributed to the constant availability of an AI pair programmer, which offers several important advantages:

• Code generator: The AI co-pilot can generate boilerplate and test code, saving developers time in setting up their code projects.
• Real-time code reviews: With AI Co-Piloting, developers receive instant feedback during the development phase, allowing them to identify and address potential problems early on.
• Legacy code refactoring: The AI assistant can help developers refactor legacy code, making it more efficient, maintainable and consistent with modern best practices. This ultimately improves the overall quality of the codebase.

One of the key improvements observed with AI Co-Piloting is the ability to receive this valuable feedback without leaving the developer’s IDE, minimizing the risk of switching contexts. Context switching occurs when a developer shifts his attention from one task to another, such as searching for documentation on the Web outside his IDE.

A University of California study found that it takes an average of 23 minutes for developers to regain their concentration and productivity after switching contexts. By providing real-time support directly within the IDE, AI Co-Piloting helps developers maintain focus and workflow, further increasing their overall productivity.

AI Co-Piloting, AI-Driven Development, or whatever the name, will revolutionize software development as a tool for professionals.

We at Delta Source believe we are at the dawn of a new evolution in the software development process. AI Co-Piloting, AI-Driven Development, or whatever the name, will revolutionize software development as a tool for professionals. Using AI through a local LLM (Large Language Model) or GitHub copilot will not replace you, but a software engineer using these tools may replace you over time.

Just like when using StackOverflow, you can’t just cut and paste the suggested code, as you may unknowingly introduce a bug.

Quality software is not something you can make if you try to leave out the human aspect. As a programmer, you have to have a lot of expertise to get the best results with AI by formulating the right questions and interpreting the answers correctly. Just like when using StackOverflow, you can’t just cut and paste the suggested code, because you may unknowingly introduce a bug that way. You must always maintain control to review and adjust any suggestion made by the LLM.

You can achieve better software quality with fewer resources if you have an excellent software engineer who uses an AI Copilot tool and also understands the shortcomings. One such shortcoming is the potential release of sensitive source code when using publicly available Large Language Model service providers. There is also the ethical issue of not knowing the origin of the data used to train these models.

The solution is to have developers who are not afraid to set up their own environment with a fully locally running LLM on your laptop. This way, we keep control over the chosen model and avoid privacy issues because our own code does not leave our machines.

In the second part of our white paper “A programmer’s guide to AI,” we test how well GitHub CoPilot and a locally set up LLM score in terms of code refactoring, code reviews and unit test generation.