The Agentic AI Gap: When Tech is Used Before its Architecture is Ready - Anthony Alcaraz, Agentic AI Architect (EP 25)

Most enterprises have the technology to run agentic AI. They do not yet have the data architecture, identity layer, or empowered workforce to actually trust it. Anthony Alcaraz argues that the bottleneck for agentic AI has shifted from building the agents to building everything around them — and that the organisations most at risk are the ones keeping a human in the loop and calling it transformation. This conversation is for leaders sitting between AI pilots that worked and production systems that have not yet arrived.

💡Episode overview

Anthony joins Eva to map what changes when AI shifts from reactive systems to agents that observe, reason, and act. The conversation moves through what enterprises miss in their own data — systems of record that capture what happened but not why — and the new attack surfaces agents introduce, including tool poisoning. Anthony names the empowerment gap inside organisations: business experts who hold the knowledge agents need, with no clear path to building anything themselves. The most provocative moment lands near the end, when Anthony argues that human-in-the-loop adoption can be a way of avoiding actual transformation rather than achieving it.

🔍 Key themes discussed

• The shift from reactive to agentic systems, and what trust has to carry now
• Why most enterprise data is missing the why behind decisions
• Tool poisoning and the new attack surface for agents
• The empowerment gap between business knowledge and technical capability
• Graph architecture as the control layer for agentic reasoning
• Why human-in-the-loop can be a refusal to transform

👤 About the guest

Anthony Alcaraz works across three vantage points that rarely sit together: he architects agentic AI systems, invests in early-stage AI startups as an angel, and is the author of Agentic Graph RAG with O'Reilly. He spends most weeks in conversation with founders attempting to enter regulated enterprises, and most evenings building software with the same tools he writes about. His perspective on this episode comes from watching the same gap repeat itself across organisations of very different sizes — the technology is ready, and most of the systems around it are not.

📍 Chapter markers

• [00:00] What changes when AI moves from reactive to agentic
• [05:42] Why agents need access — and what enterprises have not built
• [10:29] The three problems: data, governance, and the people in between
• [23:13] Graph architecture and the missing why of enterprise data
• [32:06] The empowerment gap that no one has solved yet
• [45:17] In-between: where Anthony finds himself now

🔗 Links

• Anthony Alcaraz LinkedIn — https://www.linkedin.com/in/anthony-alcaraz-b80763155/
• Agentic Graph RAG (O'Reilly) — https://www.oreilly.com/library/view/agentic-graphrag/9798341623163/
• Foundation Capital context graph thesis — https://foundationcapital.com/ideas/the-case-for-context-graphs
• Related episode — Trust as an operating system in AI companions https://open.spotify.com/episode/5t4BtgevPOtMWUfB4jThWX?si=oGo2JPHNTeCTxbqkNXDJMw
• Eva Simone Lihotzky's LinkedIn: https://www.linkedin.com/in/evalihotzky/