Uploaded December 2025 | Updated September 2026, 1 hour ago

---

MCP has enabled a wave of powerful agent demos — but production systems demand more than clever tool calls.

In this talk, we explore what changes when MCP-based agents move from experimentation to real-world use inside enterprises.

The key insight:
LLMs are better at writing code than they are at repeatedly calling tools.
Architectures that lean into this reality — generating executable code, sandboxing it, and treating tools as APIs — unlock dramatic gains in efficiency, determinism, and scalability.

You’ll see:
• Why traditional tool-calling breaks down as agents grow more complex
• How code-generation reduces token usage and improves correctness
• What “real agents” look like once they leave the demo stage
• Why standards like MCP matter for long-term interoperability
• And what enterprises must solve next: identity, authorization, and governance

As agent systems begin to act on behalf of users, organizations, and workflows, enterprise identity becomes infrastructure — not an afterthought.

This is the architectural shift teams need to understand before deploying agents into production environments.