From RPA to AI Agents: Why I Left Enterprise to Build Agent Infrastructure

In 2020, Microsoft acquired Softomotive — the company where I served as Managing Director for APAC. Our RPA product became Microsoft Power Automate. It was a defining moment in enterprise automation. Five years later, I'm building something that makes RPA look like a calculator compared to a smartphone. ## The RPA Era Robotic Process Automation was the first wave of enterprise automation. We taught software robots to click buttons, fill forms, and move data between systems. It worked. Enterprises saved millions. But RPA had fundamental limitations: - Brittle — any UI change breaks the automation - Rule-based — can't handle exceptions or ambiguity - Narrow — each bot does exactly one workflow - Expensive to maintain — requires constant updates I spent 18 years in this world, working with Microsoft, Walmart, Flipkart, Axis Bank, Deloitte, and more. I understood enterprise automation deeply — both its power and its ceiling. ## The AI Agent Revolution AI agents are different in every dimension: - Adaptive — they understand intent, not just instructions - Intelligent — they reason about edge cases - General — one agent system handles many workflows - Self-improving — they learn from feedback But here's what most AI startups get wrong: they build demos, not systems. An enterprise can't trust an AI agent that: - Has no spending limits on LLM calls - Stores credentials in plaintext - Can't be audited or controlled - Has no kill switch That's why I'm building ShackleAI — the governance, security, and operational layer that enterprises need before they can trust autonomous agents. ## What I'm Building Now ShackleAI is an 11-microservice platform that provides: - LLM Gateway with multi-provider routing and PII scrubbing - Credential Vault with AES-256 encryption and auto-rotation - Agent Governance with RBAC, ABAC, and kill switches - Orchestrator for multi-agent pipeline execution - Cost tracking across 100+ LLM providers Plus open-source tools: an MCP memory server (1,440+ weekly downloads) and a Python agent framework on PyPI. ## Why This Matters The enterprise automation market is worth $20B+ and growing. But the next wave isn't RPA — it's AI agents with enterprise-grade infrastructure. I've seen both sides: the enterprise buyer who needs compliance, security, and auditability — and the AI engineer who needs flexibility, speed, and powerful abstractions. Building the bridge between those two worlds is exactly where I want to be.