The Fifth Elephant 2026 Winter Edition

Submission type:
Showcase - new ideas - 20 mins

Describe your session in 2 paragraphs:

AI agents are increasingly capable of reasoning over information and operating digital tools, but the problem changes fundamentally when an agent needs to interact with the physical world. Physical environments introduce sensors, heterogeneous devices, legacy infrastructure, unreliable state, communication protocols, latency, safety constraints and feedback loops. An agent that can generate the correct instruction is not necessarily an agent that can safely and reliably execute that instruction in the real world.

MacroVision Claw is our work-in-progress attempt to build an edge-native Physical AI system that connects AI agents with existing physical infrastructure. It is designed to let agents perceive physical state, reason about it, interact with heterogeneous devices and verify the outcome of their actions across systems such as HVAC, lighting, pumps and energy infrastructure. In this 20-minute showcase, we will walk through the architecture, demonstrate the approach we are developing, and share the engineering questions and constraints we have encountered while moving AI agents from digital reasoning toward physical action.

Mention 1-2 takeaways from your session:

1. Physical AI introduces a different class of engineering problems: physical state, uncertainty, actuation, feedback and safety become first-class concerns for AI agents.

2. An edge-native device abstraction layer can potentially allow AI agents to work with heterogeneous existing infrastructure without requiring every physical system to be rebuilt for AI.

Which audiences will benefit most from your session?

AI/ML engineers, AI-agent builders, edge-AI practitioners, IoT and industrial automation engineers, robotics practitioners, and researchers interested in Physical AI and deploying AI into real-world environments.

Add your bio: who you are, where you work, and any relevant context:

Shaurya Srivastava is the Founder & CEO of MacroVision AI, an AI systems company based in India. He has 16+ years of experience across AI, technology, global sourcing and supply chains, new product development and emerging technologies. At MacroVision AI, he works on AI-native systems spanning Physical AI, AI infrastructure and quantum AI. His current work focuses on building AI agents that can move beyond digital interfaces and interact with real-world infrastructure.

What don’t you know yet, or what do you want help with?

We are still exploring how much autonomy a Physical AI agent should have when it can affect real infrastructure, how physical state and uncertainty should be represented to an agent, and where safety constraints should live across the agent, edge and device layers.

We would particularly value feedback on agent evaluation, edge AI, device abstraction, industrial automation, physical-world safety and methods for verifying that an action requested by an AI agent actually occurred in the physical environment.

Match-making — I can help with:

Ideas
Experience
Technique
Critique

Match-making — I need help with:

Subject Matter Experts
Experience
Technique
Collaborator
Critique

Match-making — I’d like to meet:

Researchers
Practitioners
Domain experts
Tool builders

Topic/domain tags:

AI agents
Physical AI
Edge AI
Safety