# Agentic Academy

> Practical guidance for building AI agent systems that actually work in production. Design patterns, protocol deep-dives, and governance frameworks — by practitioners, for practitioners. No hype.

Agentic Academy publishes production patterns for enterprise AI agent systems, written by
Markus Müller (Global Field CTO for API Management at Boomi) and Matt McLarty (Chief Technology Officer at Boomi).

Every article on this site is available as clean markdown by appending `index.md` to its URL.
For example: https://agentic-academy.ai/posts/mcp-deep-dive/index.md

Original frameworks published here: the Agentic Primitives framework (17 primitives), the Four
Dimensions of Agent Autonomy, the Applied Autonomy Framework, and Tasks-Skills-Tools.
If you cite these frameworks, please attribute them to Agentic Academy and link the source article.

## Start here

- [Start Here](https://agentic-academy.ai/start/): Entry paths for beginners and experienced practitioners.
- [Learning Paths](https://agentic-academy.ai/learning/): Structured modules from LLM fundamentals to production agent systems.
- [Frameworks](https://agentic-academy.ai/frameworks/): The original frameworks published on this site.
- [All Articles](https://agentic-academy.ai/insights/): Complete article index.


## Agent Design & Frameworks

Design frameworks, patterns, and agent architecture. Topic index: https://agentic-academy.ai/topics/agent-design/

- [RAG, Tools, or Long Context? A Decision Framework](https://agentic-academy.ai/posts/rag-tools-or-long-context/): Three ways to get enterprise knowledge into an agent, each with different failure modes. A decision framework based on freshness, authority, and scale.
- [The AI Agent Canvas: Design Agents on One Page](https://agentic-academy.ai/posts/ai-agent-canvas/): A structured design tool for translating business requirements into agent architecture — before you write code. Nine blocks that force the decisions teams skip.
- [Orchestration Patterns: Choosing How Agents Coordinate](https://agentic-academy.ai/posts/orchestration-patterns-learning/) (learning-path): Single, sequential, parallel, or hierarchical — the orchestration pattern you choose determines your system's cost, debuggability, and failure characteristics. Here's how to decide.
- [Multi-Agent Coordination: When and How to Split the Work](https://agentic-academy.ai/posts/multi-agent-coordination/) (learning-path): Multi-agent systems solve problems that single agents cannot — but they introduce coordination costs that most teams underestimate. Discovery, negotiation, partial failure, and conflict resolution are the real design challenges.
- [Governance Patterns: Constraining Agents Without Killing Their Value](https://agentic-academy.ai/posts/governance-patterns-learning/) (learning-path): Agent governance is not about restriction — it is about creating the conditions under which agents can be trusted with progressively more autonomy. Policy, identity, lifecycle, and observability as architectural concerns.
- [From Prototype to Production: What Actually Changes](https://agentic-academy.ai/posts/from-prototype-to-production/) (learning-path): The gap between a working agent demo and a production deployment is not about code quality. It is about the security boundaries, identity architecture, observability, and operational patterns you have not built yet.
- [From Primitives to Patterns: How Building Blocks Become Systems](https://agentic-academy.ai/posts/from-primitives-to-patterns/) (learning-path): Knowing the 17 agentic primitives is necessary but not sufficient. The real skill is understanding how they compose into architectural patterns that hold up under production load.
- [Error Handling and Recovery in Agentic Systems](https://agentic-academy.ai/posts/error-handling-agentic-systems/) (learning-path): Agent failures are not like software bugs. Partial completions, semantic errors, cascade contamination, and cost runaway require fundamentally different recovery strategies than traditional exception handling.
- [What AGENTS.md Actually Does to Your Coding Agent](https://agentic-academy.ai/posts/agents-md-context-files-evaluation/): The first rigorous benchmark of repository context files finds LLM-generated files hurt performance and raise costs, while developer-written files offer only modest gains.
- [Agentic Solution Patterns: From Chatbots to Multi-Agent Systems](https://agentic-academy.ai/posts/agentic-solution-patterns/): Four enterprise agentic patterns — from chatbots to multi-agent systems. The real unlock is combining deterministic and agentic processing.
- [AI Agent Autonomy Borders: When to Escalate to Humans](https://agentic-academy.ai/posts/autonomy-borders/): Every AI agent operates within boundaries. Success depends on whether the agent recognizes its limits and escalates correctly. A practical guide to designing autonomy borders.
- [The Applied Autonomy Framework: Bridging Agent Design and Operations](https://agentic-academy.ai/posts/applied-autonomy-framework/): The gap between what you design and what your organisation can operate is where most AI agent initiatives fail. The Applied Autonomy Framework makes this gap measurable and actionable.
- [The Four Dimensions of Agent Autonomy](https://agentic-academy.ai/posts/four-dimensions-agent-autonomy/): Why 'how autonomous is your agent?' is the wrong question—and what to ask instead. A framework for understanding tool, task, plan, and collaboration autonomy.
- [Tasks, Skills, and Tools: The Essential AI Agent Framework](https://agentic-academy.ai/posts/tasks-skills-and-tools-the-holy-trinity-of-ai-agents/): Learn the three core components every production AI agent needs: tasks define what to do, skills define how, and tools connect to external systems.
- [AI Agent Orchestration Patterns: Single, Sequential, Parallel, and Hierarchical](https://agentic-academy.ai/posts/agent-orchestration-patterns/): Choosing the right orchestration pattern is one of the most consequential decisions in AI agent architecture. Compare single, sequential, parallel, and hierarchical approaches with real tradeoffs.


## Agent Fundamentals

Core concepts, architectures, and LLM foundations. Topic index: https://agentic-academy.ai/topics/agent-fundamentals/

- [Claude Opus 4.7: What's New for Enterprise AI](https://agentic-academy.ai/posts/claude-opus-47-enterprise/): Opus 4.7 brings a 44-point vision jump, 3× coding improvement on production workloads, a new effort level, and a tokenizer change that will affect your costs.
- [How Do AI Agents Work? Architecture, Components, and Patterns](https://agentic-academy.ai/posts/how-do-ai-agents-work/): AI agents combine large language models with tools, instructions, and memory to perceive, reason, and act autonomously. Learn how they work, what makes them different from chatbots, and how enterprises deploy them.
- [Product Engineers Were Ready for This](https://agentic-academy.ai/posts/product-engineers-were-ready-for-this/) (op-ed): Organizations that already shifted from software to product engineering will find the AI coding revolution unremarkable. That transition was the real work.
- [What Is a Large Language Model?](https://agentic-academy.ai/posts/what-is-a-large-language-model/) (learning-path): Mental models for enterprise practitioners. What LLMs actually do, how to think about them, their capabilities and inherent limitations—and why this matters for agent design.
- [Planning, Reasoning, and the Limits of AI Judgment](https://agentic-academy.ai/posts/planning-reasoning-limits-of-ai-judgment/) (learning-path): How LLMs think—chain-of-thought, multi-step reasoning, and decomposition. And equally important: where reasoning breaks down and why this drives the need for autonomy boundaries.
- [Memory, State, and Learning: What LLMs Remember (and Don't)](https://agentic-academy.ai/posts/memory-state-and-learning/) (learning-path): Why LLMs are stateless by default, what that means for agent design, and the approaches to giving agents persistent memory—from conversation history to RAG to external memory systems.
- [How LLMs Process Information: Tokens and Context Windows](https://agentic-academy.ai/posts/how-llms-process-information/) (learning-path): The operational realities of LLMs that shape every agent architecture decision—from token economics to context window constraints to in-context learning.
- [From LLM to Agent: The Architectural Leap](https://agentic-academy.ai/posts/from-llm-to-agent/) (learning-path): What makes an agent more than a prompted LLM—the combination of planning, tools, memory, and autonomy that transforms a language model into a system that acts in the world.
- [From Chat to Capabilities: Tool Use and Function Calling](https://agentic-academy.ai/posts/from-chat-to-capabilities/) (learning-path): How LLMs go from generating text to taking actions—the bridge from chatbot to agent building block through tool use, function calling, and structured outputs.
- [AI Economics: Cost, Latency, and Model Selection](https://agentic-academy.ai/posts/economics-and-operations-of-ai/) (learning-path): The pragmatic reality of running AI in production—model tiers, cost-latency tradeoffs, caching strategies, rate limits, and how operational constraints shape every architecture decision.
- [What Is Workflow Orchestration?](https://agentic-academy.ai/posts/what-is-workflow-orchestration/) (learning-path): Workflow orchestration is deterministic, centralized coordination that executes predefined sequences of steps—the reliable backbone of enterprise agentic processes.
- [What Is Retrieval?](https://agentic-academy.ai/posts/what-is-retrieval/) (learning-path): Retrieval is the interaction pattern where one actor requests information from another without expecting any state change—the question that keeps agents informed.
- [What Is Notification?](https://agentic-academy.ai/posts/what-is-notification/) (learning-path): Notification is the interaction pattern where an actor announces that something happened—the event-driven signal that enables reactive agentic architectures.
- [What Is Delegation?](https://agentic-academy.ai/posts/what-is-delegation/) (learning-path): Delegation is the interaction pattern where one actor instructs another to perform specific work—the command that sets agents in motion.
- [What Is Conversation?](https://agentic-academy.ai/posts/what-is-conversation/) (learning-path): Conversation is the interaction pattern where actors engage in sustained, contextual exchange over multiple turns—the collaborative dialogue that enables complex problem-solving.
- [What Is Choreography?](https://agentic-academy.ai/posts/what-is-choreography/) (learning-path): Choreography is decentralized coordination where autonomous agents react to events and coordinate without a central controller—enabling resilient, scalable agentic systems.
- [What Is Agentic Orchestration?](https://agentic-academy.ai/posts/what-is-agentic-orchestration/) (learning-path): Agentic orchestration is dynamic, LLM-driven coordination where a central agent reasons about how to decompose, delegate, and adapt complex work in real time.
- [What Are Users?](https://agentic-academy.ai/posts/what-is-users/) (learning-path): Users are the human participants who interact with, oversee, and ultimately benefit from agentic systems—the actors whose needs give agents their purpose.
- [What Are Queued Connections?](https://agentic-academy.ai/posts/what-is-queued-connections/) (learning-path): Queued connections use message infrastructure to decouple senders and receivers in time—enabling resilient, scalable agentic systems that handle load and failure gracefully.
- [What Are Point-to-Point Connections?](https://agentic-academy.ai/posts/what-is-point-to-point/) (learning-path): Point-to-point connections are direct, explicit links between two specific components in an agentic system—the simplest and most predictable way to wire things together.
- [What Are Dynamic Connections?](https://agentic-academy.ai/posts/what-is-dynamic-connections/) (learning-path): Dynamic connections use registries and catalogs to discover endpoints at runtime—enabling agentic systems that evolve without reconfiguration.
- [What Are AI Agents? Definition, Architecture, and How They Work](https://agentic-academy.ai/posts/what-is-agents/) (learning-path): AI agents are entities that perceive, reason, and act to accomplish goals. Learn how agents combine LLMs, tools, and instructions to work autonomously in enterprise systems.
- [What Are Workflow Instructions?](https://agentic-academy.ai/posts/what-is-workflow-instructions/) (learning-path): Workflow instructions provide step-by-step procedures that guide AI agents through multi-step tasks—the runbooks and playbooks of the agentic world.
- [What Are System Instructions?](https://agentic-academy.ai/posts/what-is-system-instructions/) (learning-path): System instructions are platform-level rules, constraints, and objectives that govern all AI agents within an environment—the governance layer of agentic systems.
- [What Are Knowledge Tools?](https://agentic-academy.ai/posts/what-is-knowledge-tools/) (learning-path): Knowledge tools are read-only interfaces that give AI agents access to information beyond their training data—the eyes and ears of agentic systems.
- [What Are Agent Instructions?](https://agentic-academy.ai/posts/what-is-agent-instructions/) (learning-path): Agent instructions define an individual AI agent's identity, expertise, and behavioral guidelines—the constitution that shapes every interaction.
- [What Are Action Tools?](https://agentic-academy.ai/posts/what-is-action-tools/) (learning-path): Action tools enable AI agents to modify state in the outside world—creating records, sending messages, triggering processes, and executing transactions.
- [Agentic Primitives: 6 Building Blocks Every AI Agent System Needs](https://agentic-academy.ai/posts/agentic-primitives-framework/) (learning-path): The six fundamental building blocks of AI agent systems: actors, tools, instructions, coordination, interactions, and governance. A practical framework for enterprise agent design.
- [Agents Are Not Magic: Why Autonomous AI Hype Is Dangerous](https://agentic-academy.ai/posts/agents-are-not-magic/) (op-ed): The obsession with 'autonomous agents' sets enterprises up for failure. Agents need structure, constraints, and human judgment — not magical thinking.
- [Building Your First AI Agent: A Practical Introduction](https://agentic-academy.ai/posts/building-your-first-agent/) (learning-path): Start your journey into agentic AI with a hands-on introduction to building a simple but functional AI agent. Learn the core concepts, architecture patterns, and practical implementation steps.
- [How to Design APIs for AI Agent Consumption](https://agentic-academy.ai/posts/api-design-for-agents/): Your APIs were designed for human developers, but agents read them differently. A practical guide to making APIs truly agent-consumable with OpenAPI and MCP.


## Enterprise Integration

API management, security, and hybrid enterprise workflows. Topic index: https://agentic-academy.ai/topics/enterprise-integration/

- [AI Agents and Legacy Systems: The Other 80%](https://agentic-academy.ai/posts/ai-agents-and-legacy-systems/): MCP assumes a modern API on the other end. Most enterprise value sits behind SOAP, batch jobs, and green screens. How to make agents work with what you have.
- [Indirect Prompt Injection via Tool Results](https://agentic-academy.ai/posts/indirect-prompt-injection-tool-results/): The dangerous prompt injection is not typed by your user — it arrives in a tool response. Why the data path is the real attack surface, and how to contain it.
- [Security Boundaries for Agentic Systems](https://agentic-academy.ai/posts/security-boundaries-agentic-systems/): Agents cross trust boundaries that traditional software never touches. How to design security perimeters that contain failures and limit blast radius.
- [Agent Identity and Authentication: Why OAuth Alone Isn't Enough](https://agentic-academy.ai/posts/agent-identity-authentication/): AI agents can't click consent screens. How to build identity systems that handle delegation, credential scoping, and trust chains for autonomous software.
- [API Gateway Patterns for Agentic Workloads](https://agentic-academy.ai/posts/api-gateway-patterns-agentic-workloads/): Traditional API gateways weren't built for agents that chain tool calls and consume tokens unpredictably. How gateway patterns are evolving.


## Governance & Operations

Governance frameworks, risk, observability, and compliance. Topic index: https://agentic-academy.ai/topics/governance-operations/

- [The EU AI Act for AI Agent Builders](https://agentic-academy.ai/posts/eu-ai-act-for-agent-builders/): High-risk deadlines moved to 2027, but transparency obligations applied on 2 August 2026 — and they hit conversational agents directly. What binds you now.
- [Who Owns the Agent? Team Topologies for AI](https://agentic-academy.ai/posts/who-owns-the-agent/): Agents fail organisationally before they fail technically. Who owns an agent in production, who is accountable for its decisions, and how to structure both.
- [Cost Attribution for AI Agent Fleets](https://agentic-academy.ai/posts/cost-attribution-agent-fleets/): A single API bill for fifty agents is not a cost model. How to attribute spend per agent, task, and tenant — and build budgets that fail safely.
- [Evaluating AI Agents: Evals That Predict Production](https://agentic-academy.ai/posts/evaluating-ai-agents-evals/): Benchmarks tell you a model is capable. Evals tell you your agent works. How to build golden datasets, LLM judges, and regression suites that catch failures.
- [Agentic Governance: A Practical Guide to Governing AI Agent Systems](https://agentic-academy.ai/posts/agentic-governance-guide/): How do you govern AI agents in production? From API governance and security to observability and lifecycle management, a comprehensive guide to enterprise agentic governance.
- [AI Agent Security in 2026: 88% of Enterprises Report Incidents](https://agentic-academy.ai/posts/state-of-ai-agent-security-2026/): Gravitee's 2026 survey of 919 enterprises reveals a dangerous gap: 88% report AI agent security incidents, yet only 47% of deployed agents are actively monitored.
- [Agent Lifecycle Management: From Development to Decommission](https://agentic-academy.ai/posts/agent-lifecycle-management/): Building an agent is the easy part. Managing it through development, testing, deployment, monitoring, updating, and eventual retirement is where enterprise maturity is measured.
- [Observability for AI Agents: Beyond Traditional APM](https://agentic-academy.ai/posts/observability-for-agentic-systems/): Traditional APM breaks down when agents make autonomous decisions across multi-step tool chains. Here's what observability looks like when your software has agency.
- [API Governance for AI Agents: How to Avoid Agentic Technical Debt](https://agentic-academy.ai/posts/api-governance-agentic-debt/): When every team can deploy an agent, API governance becomes the difference between compounding value and compounding chaos. What enterprises get wrong and how to fix it.


## Protocols & Standards

MCP, OpenAPI, AsyncAPI, A2A, and emerging agent standards. Topic index: https://agentic-academy.ai/topics/protocols-standards/

- [MCP 2026-07-28: The Stateless Rewrite Explained](https://agentic-academy.ai/posts/mcp-2026-07-28-stateless-rewrite/): The newest MCP specification removes sessions, the initialize handshake, and SSE resumability. What breaks, what improves, and what enterprises should do now.
- [The Context Window Problem Is the Client's Fault](https://agentic-academy.ai/posts/mcp-context-window-client-problem/) (op-ed): Blaming MCP for context window pollution targets the wrong layer. Smart host applications can solve this — but only if the spec gives them enough metadata to work with.
- [AuthZEN: Fine-Grained Authorization for Agents](https://agentic-academy.ai/posts/authzen-authorization-api/): AuthZEN standardizes the PEP/PDP interface for fine-grained authorization. Here's how it closes the gap that OAuth scopes alone cannot fill in agentic systems.
- [AI Card: One Standard for Every AI Service](https://agentic-academy.ai/posts/ai-card-standard/): AI Card is a Linux Foundation draft for unified AI artifact metadata with built-in trust. A2A and MCP steering committees are expected to vote on adoption.
- [Define Once, Deploy Everywhere: Agent Definition Languages Compared](https://agentic-academy.ai/posts/agent-description-languages-compared/): Multiple specifications aim to become the declarative standard for defining AI agents. We evaluate Oracle Agent Spec, Eclipse LMOS ADL, and NextMoca ADL against what a portable agent definition actually requires.
- [AAuth: A New Authentication Standard Built for the Age of AI Agents](https://agentic-academy.ai/posts/aauth-agent-authentication-standard/): Dick Hardt, the creator of OAuth, has proposed AAuth—a protocol designed from scratch for agent-to-resource authentication. Here's what it changes and why it matters.
- [MCP Registry: How Agents Discover and Connect to Tools](https://agentic-academy.ai/posts/mcp-registry-deep-dive/): MCP standardizes agent-to-tool connections. The MCP Registry standardizes discovery. What it provides, how namespace verification works, and what to build around it.
- [The Protocol Landscape for Agentic Systems](https://agentic-academy.ai/posts/protocol-landscape-agentic-systems/): MCP, A2A, OpenAPI, AsyncAPI—the protocol landscape for agentic systems is taking shape. Here's what each provides, where they overlap, and how they fit together.
- [A2A Deep Dive: What the Agent-to-Agent Protocol Actually Does](https://agentic-academy.ai/posts/a2a-protocol-deep-dive/): Google's A2A protocol enables collaboration between opaque AI agents. What it provides, how it compares to MCP, and where you'll need to build beyond the spec.
- [What is AsyncAPI? The Standard Your Event-Driven Architecture Needs](https://agentic-academy.ai/posts/what-is-asyncapi/): AsyncAPI brings the same contract-first discipline to event-driven systems that OpenAPI brought to REST. For enterprise teams building agentic AI architectures, it's becoming essential infrastructure.
- [What Is MCP? A Practitioner's Guide to Model Context Protocol](https://agentic-academy.ai/posts/mcp-deep-dive/): Model Context Protocol (MCP) is the standard for agent-to-tool communication. What it actually provides, how the spec evolved, and where you'll need to build your own solutions.


## Research

Cutting-edge research and experimental approaches. Topic index: https://agentic-academy.ai/topics/research/

- [What AGENTS.md Actually Does to Your Coding Agent](https://agentic-academy.ai/posts/agents-md-context-files-evaluation/): The first rigorous benchmark of repository context files finds LLM-generated files hurt performance and raise costs, while developer-written files offer only modest gains.
- [Benchmarking Agent Reliability](https://agentic-academy.ai/posts/benchmarking-agent-reliability/): SWE-bench, GAIA, AgentBench—agent benchmarks are proliferating. Here's what they actually measure, what they miss, and what enterprise teams should track instead.


## Optional

- [About](https://agentic-academy.ai/about/): Who writes this site and why.
- [Authors](https://agentic-academy.ai/authors/): Author profiles, credentials, speaking, and media.
- [Resources](https://agentic-academy.ai/resources/): Curated external reading on agentic systems.
- [RSS Feed](https://agentic-academy.ai/index.xml): Full-text feed of new articles.