As one of the world’s largest semiconductor companies, AMD is making bold moves to incorporate artificial intelligence (AI) into its software development lifecycle. AMD is secretive as it is focused—AMD employs nearly half its workforce as software engineers. Even more importantly, they deeply understand the transformative potential of generative AI technologies that have captured the world’s imagination in recent years. AMD is taking a more holistic approach with AI coding copilots. This ongoing public-private partnership aims to increase the productivity of scientific research while protecting associated intellectual property.
The company developed and has implemented a range of AI tools that help developers do their jobs better and faster every day. These tools support learning, explanation, code generation, code review, test generation, triage and debugging. Despite the benefits, AMD is equally focused on the limitations and risks inherent with AI. The company is quick to remind customers that AI isn’t always right. As such, it focuses a lot more on the ethical and responsible usage and implementation of these technologies.
The Role of AI in Software Development
At AMD, we leverage AI tools like the world’s smartest autocomplete. They can recommend additional lines or even complete functions of code to help developers maximize efficiency. These tools tremendously increase the productivity of professional software engineers. They do this through the application of deep convolutional neural networks (CNNs) trained on a dataset of more than 20,000 “golden” images and 2,000 distorted images. The development and test phases are the most resource-intensive. AMD has 60 engineers working on each phase. Despite the heavy investment in these areas, developing and testing new code constitutes only about 40 percent of a developer’s workload.
The rest of a software engineer’s day goes to all the other priorities. This includes learning new technologies (10 percent), triaging and debugging problems (20 percent), reviewing and optimizing code (20 percent), and documenting code (10 percent). With a firm focus on AI, AMD is looking to disrupt these processes and get candidates hired at a drastically faster rate.
AMD has fully adopted AI into the functional testing phase. As a direct result, they’ve reduced manual test effort by 15 percent. Plus, the scenarios it can test has surged by 20 percent. This new feature helps speed up the overall software testing cycle while ensuring higher quality software.
Intellectual Property Considerations
For all the promising capabilities AI tools have, AMD is cautious when it comes to their implementation. The implications are significant enough that the company is deeply concerned regarding potentially infringing third-party intellectual property rights when using AI-generated recommendations. There are worries about the unintended release of its own intellectual property as it interacts with AI tools that are publicly available.
To address these risks, AMD is committed to the secure and responsible use of AI technologies. The firm remains dedicated to ensuring compliance with existing legal frameworks and promoting innovation by integrating AI responsibly.
Future Expectations and Productivity Gains
AMD is doing a lot to educate developers on using and integrating AI tools into the developer’s environment. They project this change will increase their overall team productivity by over 25 percent. This is a logical increase as the company continues to iterate on their processes and improve workflows through more advanced AI technology.
AMD typically sticks to an aggressive product release cadence of around six months. This continuous-deployment cycle holds true for new software packages—including AMD’s Adrenalin graphics-software package. Following that, the team goes into a support phase that takes place over an additional three to six months. Meanwhile, for twelve months, twenty engineers are on full-time support, ten engineers are designing, and five create definitions. Whether AI supports this whole cycle or not, it has the potential to greatly shorten timelines and improve convergence product quality.