How to accelerate software development with AI

We’ve been chatting a lot about AI here recently. I tried to argue against prophecies of AI apocalypse without disregarding concrete issues such as model bias or LLMs ability to generate plausible-sounding lies. AI could enable new generations of builders and trigger a wave of new careers that are as unimaginable for us today as my job as a mobile platform engineer was for my grandparents, but only if we take it seriously, without either fear or hype.

Let’s shift gears and move from the abstract to the practical.

My article in GitHub’s The ReadME Project, Accelerate test-driven development with AI, takes a look at the kind of synergy that AI brings to software development today.

The article shows how to Copilot enhances the Red-Green-Refactor loop at the core of Test-Driven Development by reducing the lag between you picturing a change in your mind and the code to implement it appearing on screen. It shows how to write prompts that can generate entire test doubles or how to guide Copilot to refactor your code.

Here’s one of my favorite examples:

Reading through the guide, you’ll also notice how, at no point, Copilot is ever in charge. Like the artificial intern that it is, GitHub’s AI can only suggest code. It’s up to you, the human, to decide whether to use the suggestion and to guide Copilot with precise prompts.

Whether you are interested in software development or not, I hope you’ll find the article interesting as a hands-on look at the capabilities and limitations of LLMs.

I encourage you to give integrating generative AIs in your workflow a shot. Once you find the combination that is right for you, I guarantee they’ll provide powerful leverages.