Telecom operators are moving beyond conversational AI towards AI that can take real action across live telecom systems. However, achieving production-grade Agentic AI faces two barriers: governance and comprehensibility. Responsible telcos want to give AI agents a safe, auditable way to execute actions on BSS systems with the right authorisation, policy control and subscriber oversight. To achieve this, AI must understand telecom capabilities in ways that exceed current API definitions which expose interface structure and parameters, but not intent, operational meaning or usage constraints. As a result, AI-to-BSS integration requires significant deployment effort and remains in bespoke, limiting reuse and slowing adoption. The Essential Framework for Telecom Agentic AI addresses both challenges through a governed execution framework grounded in TM Forum Open Digital Architecture (ODA) and the AI-Native Blueprint. We seek to establish a unified architecture that transforms declarative telecom capabilities into governed, AI-executable services, enabling safe and scalable agent interaction. A core innovation is to transform declarative ODA component capabilities into governed, runtime-executable tools via the Canvas AI Operator, establishing a lifecycle from design-time definition to safe AI execution. This enables agents to interact with capabilities rather than APIs, combining policy-governed execution, structured business context, semantic capability understanding, and human oversight. One of this project’s core contributions is the extension of ODA Canvas component definitions with agenticCapabilities . The agenticCapabilities add the semantic and governance metadata required to make BSS capabilities AI-comprehensible and AI-callable, allowing AI agents to discover, govern and reuse ODA-aligned components without bespoke integration for each use case. The framework also introduces a self-improving telecom Knowledge Graph, enabling recommendations to continuously improve based on real operational outcomes. We demonstrate the framework through representative customer care and AI-driven operations scenarios, showing how AI agents retrieve the right context, recommend the right actions, and execute safely within policy-defined boundaries. The project establishes a reusable, standards-aligned reference architecture that enables operators to move from isolated AI pilots to scalable, governed operational AI, while contributing practical implementation insights to TM Forum on the operationalization of Agentic AI within ODA-aligned environments.