How enterprises can ride the AI coding wave

While artificial intelligence (AI)-powered coding assistants have the potential to transform software development, lingering concerns about AI replacing software engineers remain.

Chang Sau Sheong, chief technology officer and deputy chief executive of the Government Technology Agency of Singapore (GovTech), addressed these concerns at this year’s Stack developer conference, urging developers to prioritise value, security and compliance when implementing AI coding tools in enterprises.

Chang traced the evolution of coding assistants from basic spell checkers to today’s sophisticated AI-powered tools, highlighting the inflection point marked by GitHub Copilot and ChatGPT. “We all know that software development is changing,” he said. “AI has and continues to transform our work,” he added, noting that a recent Stack Overflow survey revealed 76% of developers are already using or plan to use AI coding tools.

However, enterprise adoption requires a more considered approach than individual use, particularly around value and return-on-investment. “Many claim AI coding tools deliver work 50% faster, but we need to be careful about what that means,” said Chang. “What kind of work? If you claim a 50% gain, can the CEO take back 50% of your budget?”

Noting that software development goes beyond coding to include debugging, refactoring and testing, he called for a nuanced understanding of how AI coding tools can benefit and fit into development workflows. “We need to provide the correct message to stakeholders, beyond time savings,” said Chang.

Security is also critical, he said, cautioning against data breaches, intellectual property leaks and code vulnerabilities introduced by AI assistants. “Imagine using an AI assistant massively, and if there’s data leakage, your competitors will know what products you’re building,” warned Chang, adding that enterprises must consider the hosting environment, data sensitivity and potential attack vectors.

Chang also advised developers to evaluate the security implications of various deployment options, including software-as-a-service (SaaS) offerings such as ChatGPT, cloud-based models such as Amazon Q Developer, or models such as Meta’s Llama, which can be hosted on-premise. He emphasised the need to assess risks thoroughly, taking into account the specific applications being developed and sensitivity of the data involved.

Compliance is another key consideration. Chang highlighted the importance of navigating the regulatory landscape, particularly for sectors such as finance, healthcare and government. “Your current policies might not be compatible for using AI coding tools,” he said, urging enterprises to review and adapt policies to address intellectual property concerns and other challenges posed by AI coding assistants.

At the conference, Chang also shared GovTech’s experience with a pilot programme involving GitHub Copilot and GitLab, focusing on key outcomes such as developer well-being, productivity, and collaboration. The pilot involved 70 participants over four months, demonstrating 24% average productivity improvement.

GovTech is now rolling out AI coding tools to its wider developer community, and is looking to do so for other government agencies. It has also revised its policies to support the use of AI coding tools, including different hosting options. “It’s been a journey, and while it can be a bit scary, I find it really exciting, and I can’t wait to see how much more we can do,” said Chang.