AMD's Dr. Su in Conversation with Kai-Fu Lee: AI Transformation Must Be CEO-Driven, Future "DRI" (Directly Responsible Individual) Will Be the Core of Enterprises | Live from the Event

TL;DR

• AMD CEO Lisa Su and 01.AI founder Kai-Fu Lee identified 2026 as the pivotal year for the shift from generative AI to Agentic AI, enabling autonomous multi-agent systems to manage entire enterprise workflows.
• Lee argues that AI transformation must be CEO-driven rather than CTO-driven, focusing on structural business changes that impact financial statements rather than superficial software add-ons.
• The concept of the "DRI" (Directly Responsible Individual) is emerging as a core enterprise role, where one person orchestrates multiple specialized agents to achieve end-to-end business results.
• Lee advocates for "small models" and local-first edge computing to meet the latency and data sovereignty requirements of multi-agent architectures, aligning with AMD's hardware strategy.
• AMD's Shanghai R&D center is highlighted as one of its largest globally, reinforcing the company's commitment to the Chinese AI ecosystem and open-source innovation.

Why It Matters

This dialogue shifts the AI discourse from model capability to organizational execution, highlighting that the next competitive advantage lies in multi-agent orchestration and structural business transformation. For AI practitioners, it signals that the demand for large, general-purpose models may be surpassed by the need for efficient, specialized small models capable of local, low-latency inference. For industry leaders, it provides a clear framework for evaluating AI ROI: if deployment does not alter key financial metrics, it remains an experimental lab exercise rather than a strategic transformation.

Key Data

• User Growth Projection: Global active AI users currently exceed 1 billion and are projected to surpass 5 billion in the coming years.
• Latency Requirement: Multi-agent architectures with real-time feedback require latency of less than 100 milliseconds, driving local-first edge processing.
• Specific Use Case: Kai-Fu Lee utilized 19 distinct agents to build a personal AI assistant that aggregates company data, meeting records, and discussions for executive visibility.
• Historical Timeline: AMD has been present in China for over 30 years, with the Shanghai R&D center participating in chip design, AI software, and platform engineering.
• Paradigm Shift Years: Lee defines 2024 as the era of "AI completing a task," 2025 as "AI completing a workflow," and 2026 as "AI running an enterprise function."

Technical Details

• Multi-Agent Architecture: The system relies on specialized agents handling planning, review, execution, and risk control. These agents operate in a coordinated team structure, debating and promoting one another to surpass the capability ceiling of any single model.
• Inference-First Economy: The "agentic economy" is fundamentally an inference economy rather than a training one. This requires parallel processing of queries across multiple agents, necessitating extreme token efficiency and local processing to meet sub-100ms latency constraints.
• Small Model Strategy: 01.AI develops task-specific small models instead of relying on massive general-purpose models. This approach allows for efficient operation on specialized all-in-one hardware, ensuring data remains within the enterprise perimeter for security and sovereignty.
• DRI Operational Model: The Directly Responsible Individual acts as the central orchestrator of a network of agents. The DRI applies engineering rigor to business outcomes, monitoring not just code but also API latency, activation rates, conversion funnels, and revenue impact.

Industry Insight

• CEO Accountability: Enterprises must restructure leadership to ensure AI transformation is a top-down strategic imperative. CTOs should focus on secure deployment, while CEOs must own the structural changes to revenue and profit models.
• Hardware-Software Co-design: The rise of agentic AI creates a direct demand for hardware that supports low-latency, local inference. Companies like AMD that offer integrated AI solutions are well-positioned to capture the market segment focused on data privacy and edge computing.
• Developer Role Evolution: The most valuable developers are those who can bridge engineering and business. Understanding system orchestration and agency behavior allows individual developers to achieve the output of traditional teams, creating opportunities for "one-person companies."

zational goals rather than simple prompts. They autonomously coordinate, execute, measure, and optimize in a closed-loop feedback system, effectively running enterprise functions.

Q: Why does Lee argue that single agents are insufficient?
A: Single models have limited capabilities regardless of their size. A multi-agent architecture, where agents with different specialties (planning, execution, risk control) collaborate and debate, can break through this ceiling to provide "super intelligence" solutions that a single agent cannot achieve.

Q: How does the "DRI" concept change the role of developers?
A: It shifts the developer's focus from writing code to orchestrating results. The DRI is accountable for end-to-end cross-functional outcomes, using their engineering background to guide, evaluate, and build verification flows for the agents they manage, rather than just maintaining software infrastructure.