Rukma Sen is a Google product marketing manager whose work on Vertex AI focuses on making generative AI practical, trustworthy, and accessible. She views agentic workflows as the interface through which people will increasingly use sophisticated models without needing to understand their internals.
Her background combines English literature, product storytelling, and enterprise cybersecurity. At Microsoft, she wrote about phishing defense and security education, emphasizing realistic training and the human behavior behind organizational risk. At Google, she has applied that attention to audience needs and safety to developer-facing AI products.
How she thinks about agents
- Agents need reasoning, action, and memory. Sen defines useful agents through three interacting components: a model that reasons toward a goal, tools or external APIs that perform actions, and orchestration that maintains context and state.
- Hybrid agents balance flexibility and control. Rule-based systems can handle predictable operations while generative models explain results or respond conversationally; her example pairs structured financial analysis with customer-facing communication.
- Multi-agent supervision introduces its own failure modes. Her customer-service architecture divides work among routing, specialist, and supervisory agents. She has encountered a supervisor that repeatedly rejected another agent’s output, illustrating why evaluation criteria and stopping conditions matter.
- Responsible interaction design belongs inside the product. Sen treats privacy, cybersecurity, misinformation, and distinguishing generated content from established facts as fundamental interface concerns, particularly as AI spreads into voice systems and ambient computing.
Her AI Engineer conference presentation connects these principles to enterprise applications spanning employee support, sales, supply chains, and legal knowledge. She also applies agent workflows to marketing, using separate creation and evaluation agents with iterative human feedback.
