Building Blocks of the Agentic Enterprise

AI agents are moving from experimental systems into production: automating workflows, calling external tools, and making autonomous decisions on behalf of users and organizations. This is a fundamentally different kind of software, and it requires infrastructure that has never existed before, not because the existing stack is broken, but because it was built for software that executes predefined logic, not software that decides what to do, chooses which tools to use, and spends money on its own.

In this talk, we lay out the building blocks enterprises need before agents can operate reliably at scale:

• Model infrastructure: access to LLMs that enable reasoning and autonomy, which agents depend on to operate
• Tooling and skills:  exposing enterprise systems as structured, agent-friendly capabilities instead of human-oriented interfaces
• Memory and context: retaining and applying knowledge across tasks so agents can operate with continuity across steps
• Identity and security: agents as first-class identities (not API keys or user credentials), with scoped permissions, authentication, and auditability
• Agent runtime management: deploying, observing, and evaluating agents whose behavior is inherently non-deterministic
• Governance and boundaries: defining and enforcing constraints for systems that act beyond static application rules

Most enterprises are still building agents in isolation. The shift ahead is architectural: building the foundations that make agents reliable, observable, and controllable at scale.