# POC Prison: Why agentic systems never escape the lab and how to fix that in 90 days - Luise Freese

## Executive summary

The talk argues that most agentic AI systems fail to move from Proof-of-Concept (POC) to production because they are blocked not by model limitations, but by fundamental organizational and governance realities. The speaker proposes a structured, 90-day program focused on building the 'paved road'—an operational backbone—to ensure agents can run safely in messy, legacy enterprise environments with clear accountability.

## Key takeaways

- The POC Prison Problem: POCs are often temporary, unmeasured, and reversible experiments that fail because they lack a defined path to production ownership. This creates an 'AI zombie' state where the system exists but delivers no measurable value or transformation.
- The Three Pillars of Enterprise Readiness: Successful deployment requires addressing technical, organizational, and cultural gaps. The biggest hurdles are unclear ownership (who owns it when it breaks?), lack of dedicated funding for operations, and a culture that rewards demos over deployment frequency.
- The 90-Day Transition Program: To escape the POC prison, implement a structured 90-day program: (1) Document current reality (ugly processes/data); (2) Achieve commitment and accountability by selecting one initiative; (3) Deploy into real systems under supervision to build the paved road.
- Governance Must Be Code: Compliance and governance cannot live in slide decks or meetings. They must be embedded directly into the delivery process (e.g., 'governance as code'), making rules executable, auditable, and non-negotiable.

## Technical details

- Agent Architecture & Scope: Agents should be 'bounded agents' that own a super narrow task, operate in one system, and have measurable outcomes. Focus on delegated authority (what decisions can it make?) rather than just intelligence.
- Data Readiness: The primary technical challenge is data quality: messy operational data, inconsistent schemas, missing metadata, and unclear lineage. The solution requires establishing a centralized 'data estate' rather than relying on fragmented spreadsheets.
- Deployment & Security: Technical requirements include implementing the principle of least privilege, ensuring service principals are used for identity, maintaining comprehensive audit trails, and building robust monitoring/observability (logs, traces, tool call telemetry) into the pipeline.
- Orchestration: For multi-agent systems, implement orchestration flows with fallback logic and conflict resolution strategies to ensure safe recovery if one agent fails or is unavailable.

## Practical implications

- Shift focus from 'what AI can do' (ideation) to identifying and solving specific organizational pain points (e.g., where mistakes are frequently made).
- Mandate that governance, risk, and compliance controls are implemented *within* the delivery pipeline ('governance as code'), not reviewed afterward.
- Establish a dedicated 'tiger team' with delegated authority and protected time to bypass bureaucratic bottlenecks like endless steering committee meetings.
- Define clear decision boundaries for agents (e.g., what requires human approval) before deployment, ensuring accountability is tied to a specific person.

## Topics

Agentic AI, Enterprise Architecture, Governance, Continuous Integration/Deployment (CI/CD), Process Maturity, Data Governance

Source: https://www.youtube.com/watch?v=L-J29rWtnkY