Microsoft & Google Eye Agentic AI
Microsoft CEO Satya Nadella and Google DeepMind's Demis Hassabis are now converging on agentic AI as the next breakthrough beyond LLMs. Nadella stated the "best protection against AI-driven job displacement is to transform yourself," signaling a major push for upskilling as AI systems gain the ability to reason and execute complex tasks.
Agentic AI represents a shift from reactive models that answer prompts to proactive systems that can reason, plan, and autonomously execute multi-step tasks to achieve a goal. These agents use Large Language Models (LLMs) as a "brain" but add capabilities like memory, planning modules, and the ability to use other software tools and APIs to take action. This allows them to move beyond generating content to completing complex workflows with minimal human oversight. Microsoft is embedding agentic capabilities across its entire ecosystem, including Dynamics 365, Power Platform, and Microsoft 365. Nadella envisions an "Agentic Web" where intelligent agents collaborate to execute tasks, changing the fundamental structure of software billing from a 'per-user' to a 'per-agent' model. The company is actively promoting this shift through initiatives like its Agentic AI Research and Innovation (AARI) program and pre-built agents for sales and research. Google DeepMind is pursuing agentic AI as a core component of its path toward Artificial General Intelligence (AGI). Projects like SIMA, which trains agents across diverse 3D virtual worlds, and Gemini Robotics, which gives robots multi-step reasoning capabilities, demonstrate this focus. Hassabis has cautioned that building reliable "world models" is complex, as even small error rates can compound over long tasks, but he predicts agents will reliably complete delegated tasks within the next year. The move toward agentic AI places new demands on hardware, particularly for on-device processing. Executing complex agentic workflows locally on a device, rather than through constant cloud communication, is critical for reducing latency, ensuring user privacy, and enabling offline functionality. This trend is driving demand for more powerful and efficient custom silicon capable of handling the reasoning and planning required by autonomous agents. In manufacturing and supply chain, AI agents are already being deployed to optimize production schedules, manage inventory, and monitor supplier performance in real-time. These systems can autonomously detect quality defects, trigger maintenance orders, and even communicate with the AI agents of contractors to ensure service level agreements are met. This allows for a shift from reactive problem-solving to a more predictive and resilient operational model.
