Manish Kumar.

Agentic AI inside the enterprise security boundary

Design patterns that let AI agents handle operational work across applications, operating systems, cloud and identity while staying inside the security boundaries regulated companies require. The main ideas: cost-tiered inference routing, so model spend matches how hard each decision is; agents that compose tools across the full stack; and human approval gates built into the design from the start. These patterns run in production at one of the world's largest automakers and are described in my patent filings, papers and the Enterprise AI Architecture Handbook 2026.