Sriramprabhu Rajendran
Algorithmic Guardrails for Autonomous Machine Execution: Engineering Safety into Intelligent Agent Systems
Abstract:
The emergence of autonomous agents powered by AI marks a paradigm shift in the way intelligent machines can be used to interact with business systems. Traditional automation was based on sequential actions executed deterministically by a human operator or another system at specific intervals. Autonomous agents, however, can execute decisions instantly, trigger operations instantaneously, and string multiple workflows together without any human involvement. The following presentation will look at four architectural and algorithmic imperatives when using autonomous agents in mission-critical scenarios, as gleaned from operational experiences in distributed systems: Deterministic Idempotency Algorithms – The need for algorithmically guaranteed deduplication of retry attempts by an intelligent agent which cannot be achieved in practice with best effort methods, but requires use of intent scoped token locking and atomic state machines. Algorithmic Enforcement of Operational Bounds – Going beyond mere identity checks to impose deterministic constraints on velocity and resource usage limits and compliance that cannot be subverted by probabilistic calls from a machine client. Discovering Dynamic Capabilities Through Machine-Readable Processes — Using How Model Context Protocol (MCP) and semantic metadata agreements to allow intelligent machines to discover, analyze, and securely execute capabilities of the system at runtime without any hard-coded integration. Auditability of Structured Reasoning — Building event-based systems that record the reasoning process of autonomous machines for compliance purposes while keeping their thought process hidden and without affecting system performance. This session gives professionals practical guidelines on how to implement secure architectures that will ensure that the enterprise IT infrastructure is safe for autonomous agents. In this case, the cost of having uncontrolled decisions by the autonomous agents will be duplication of actions and system failures.
Profile:
Sriramprabhu Rajendran is an experienced IT leader and researcher with more than two decades of designing large-scale distributed systems in banking and finance. His areas of expertise include cloud-native architectures, event-driven systems, and the cutting-edge topic of Generative AI applications to enterprise infrastructures.
He is a Senior Member of IEEE and a prolific researcher contributing extensively to international literature, with articles published in the International Journal of Computer Applications (IJCA), International Journal of Engineering (IJE), and Artificial Intelligence: An International Journal (AIIJ). His research interests include scalable event-driven architectures, agentic AI orchestration frameworks, and industrial-grade GenAI deployments in regulated industries.
Serving on Technical Program Committees for IEEE international conferences such as IEEE AIIoT 2026, AMLDS 2026, and ARIIA 2026, he has done more than 80+ peer reviews in AI, cloud computing, and distributed systems.
His professional career includes heading up cross-functional engineering teams, platforms projects, and developing architectures which process millions of events every day. He has spoken at several international technology conferences such as API Days New York 2026, and thought leadership contributions through his writings for platforms like DZone, Dataversity, and Dev.to.