Descriptions:
Anush Elangovan, VP of Software at AMD, delivered a keynote at the AI Dev 26 x SF conference hosted by DeepLearning.AI, sharing how his team is experiencing and operationalizing the current wave of AI-driven software development. Drawing on the concept of the “K-shaped future” of software engineering — backed by a Harvard analysis — he argues that systems-level thinking, judgment, and problem framing are rapidly appreciating in value while syntax-level coding knowledge becomes increasingly commoditized.
Elangovan introduced the concept of “intent velocity” as the key metric leaders should optimize for: how quickly can a team move from idea to production outcome? He described running four to six autonomous agents overnight in parallel, a working pattern he says was simply impossible until recently. At AMD, this shift has enabled projects that previously seemed out of reach to become single-session efforts.
He highlighted two concrete internal AMD projects that illustrate the scale of acceleration. The first, called Geek, is an agent loop that continuously and autonomously optimizes software performance for AMD hardware customers, delivering faster token-serving speeds without manual tuning cycles. The second, Rosetta, performs on-the-fly machine ISA translation between GPU architectures — a task that historically required 200–300 engineers over four to five years. Both projects sit within AMD’s ROCm open-source software stack, which Elangovan notes is fully understood by frontier models, making AI-assisted development on AMD hardware particularly tractable.
📺 Source: DeepLearningAI · Published May 19, 2026
🏷️ Format: Keynote Launch
