The Copilot Tax: Why Your AI Strategy is Bleeding Cash

From SharePoint Developer to Power Platform Architect: Building Secure and Scalable Solutions with Michel Mendes [MVP]

In this episode of the M365 Podcast, Mirko Peters sits down with Microsoft MVP Michel Mendes to explore his remarkable journey from traditional SharePoint development to becoming a leading Power Platform Architect. Michel shares how he started his Microsoft technology career in Brazil, transitioned from C# and SharePoint development into the modern Power Platform ecosystem, and eventually moved to Ireland to continue building enterprise-grade solutions for organizations worldwide.Throughout the conversation, Michel provides valuable insights into how the Microsoft ecosystem has evolved over the years, the growing role of AI in software development, and why understanding architecture, governance, and security remains critical even in a low-code world. Whether you're a developer, solution architect, IT leader, or Power Platform enthusiast, this episode delivers practical guidance for building scalable and maintainable business applications. POWER PLATFORM EVOLUTION AND THE FUTURE OF DEVELOPMENT Michel discusses how Power Platform has transformed application development by enabling both professional developers and technically minded business users to build solutions faster than ever before. He also shares his perspective on how AI-powered development tools such as GitHub Copilot are changing the way applications are designed, prototyped, and maintained.Key topics include:• The transition from traditional development to low-code solutions • How AI is accelerating software delivery • Why developers who embrace AI will thrive • The future of Power Apps, Power Pages, and pro-code development • The importance of understanding business problems before building technology BUILDING ENTERPRISE POWER APPS THAT SCALE Creating an app is easy. Creating an app that remains maintainable, performant, and scalable for years is much harder.Michel explains the architectural principles that separate successful Power Platform implementations from those that struggle over time. He shares practical advice on designing reusable components, improving performance, and creating solutions that can grow alongside business requirements.Topics covered:• Power Apps design best practices • Building maintainable applications • Performance optimization strategies • Reusable components and architecture patterns • Measuring business value and user adoption DATAVERSE AS THE FOUNDATION OF MODERN BUSINESS APPLICATIONS A major part of the discussion focuses on Microsoft Dataverse and its role as the foundation for enterprise-grade Power Platform solutions.Michel explains why Dataverse is much more than a database and how it provides built-in governance, security, authentication, and scalability capabilities that help organizations avoid reinventing the wheel.Learn about:• Dataverse architecture fundamentals • Security and governance advantages • Building scalable business applications • Plugins versus Power Automate flows • Designing efficient data models POWER PAGES AND EXTERNAL BUSINESS SOLUTIONS Michel is widely recognized for his expertise in Power Pages, and this episode dives deep into how organizations can create secure, modern, and scalable external-facing websites powered by Dataverse.The conversation explores when Power Pages is the right choice, how it differs from Power Apps, and how recent innovations are making the platform even more attractive for professional developers.Highlights include:• Power Pages fundamentals • External portals and customer-facing applications • React and Angular-based SPA experiences • AI-assisted website development • Modern Power Pages architecture SECURITY, GOVERNANCE, AND WEB API BEST PRACTICES One of the most valuable sections of the episode focuses on security.Michel explains common mistakes developers make when exposing Dataverse data through Power Pages and outlines practical approaches for protecting sensitive information while maintaining usability.Topics include:• Dataverse table permissions • Column-level security • Power Pages Web API security • Common security vulnerabilities • Governance and compliance best practices • Penetration testing and security reviews COMMUNITY, CAREER GROWTH, AND MVP INSIGHTS Michel also shares his experiences as a Microsoft MVP and discusses the importance of contributing back to the Microsoft community through blogging, conference speaking, GitHub projects, and social media engagement.For professionals starting their Power Platform journey, he provides actionable advice on certifications, learning paths, and developing a long-term career strategy within the Microsoft ecosystem.This episode is packed with real-world experience, technical insights, and practical guidance for anyone looking to build secure, scalable, and future-ready solutions with Microsoft Power Platform.Whether you're a SharePoint veteran, a Power Platform developer, a solution architect, or simply curious about the future of low-code and AI-powered development, this conversation with Michel Mendes delivers valuable lessons from someone who has successfully navigated every stage of that journey. Become a supporter of this podcast: https://www.spreaker.com/podcast/m365-fm-modern-work-security-and-productivity-with-microsoft-365--6704921/support [https://www.spreaker.com/podcast/m365-fm-modern-work-security-and-productivity-with-microsoft-365--6704921/support?utm_source=rss&utm_medium=rss&utm_campaign=rss].

STOP BUILDING SILOED AGENTS: The Logic App Nervous System

Everyone is building AI agents.Very few organizations are building agent architectures.Across Microsoft 365, Copilot Studio, Azure OpenAI, Power Platform, and custom AI solutions, enterprises are racing to deploy copilots, bots, assistants, and autonomous workflows. Teams are creating agents for customer service, IT support, HR onboarding, knowledge discovery, incident management, and business operations.Most of them work.At least in the demo.But something very different happens when organizations move beyond a single agent and attempt to coordinate dozens of AI-powered systems across multiple business units, multiple platforms, and multiple Microsoft 365 tenants.The result is often chaos.Disconnected bots. Duplicate integrations. Credential sprawl. Governance gaps. Broken workflows. Untraceable actions. And increasingly, AI agents that cannot collaborate because they were never designed to operate as part of a larger system.In this episode, we explore why enterprise AI is repeating the same architectural mistakes organizations made during the early API revolution, why point-to-point agent integrations are becoming unsustainable, and how Azure Logic Apps is emerging as the orchestration layer that connects reasoning, execution, governance, identity, and automation into a single enterprise nervous system.If your organization is investing in Copilot Studio, Azure OpenAI, Microsoft 365 Copilot, Power Platform, or custom AI agents, this episode provides a blueprint for building agent ecosystems that actually scale. THE CHATBOT MIRAGE Most enterprise AI projects begin with a simple success story.A team creates a bot.The bot answers questions.The demo works.The project gets funded.Then another department builds another bot.And another.And another.Soon the organization has dozens of isolated AI systems solving local problems but creating enterprise-wide complexity.We explore: * Why AI demos rarely reveal architectural weaknesses * The difference between local optimization and enterprise orchestration * How siloed agents create operational debt * Why successful pilots often fail at scale * The hidden cost of disconnected automation The problem isn't the agents.The problem is the architecture beneath them. THE POINT-TO-POINT INTEGRATION TRAP Every agent needs data.Most agents get it the wrong way.Organizations frequently allow agents to connect directly to APIs, databases, SaaS platforms, and Microsoft Graph endpoints.Initially this feels efficient.Eventually it becomes unmanageable.This episode examines: * Point-to-point integration sprawl * Credential proliferation * Duplicate business logic * Decentralized error handling * Governance fragmentation * Observability challenges The more agents you deploy, the more dangerous direct integration becomes. WHY AGENTS FAIL AT ENTERPRISE SCALE The most advanced language model in the world cannot compensate for poor architecture.We discuss why: * Reasoning is not orchestration * Intelligence is not governance * Conversation is not workflow management * Tool calling is not process execution * AI is not a replacement for enterprise integration Enterprise success depends less on model sophistication and more on execution architecture. THE STATEFUL GAPOne of the most important concepts in this episode is the distinction between reasoning and memory.Most AI agents are stateless.Enterprise processes are not.We explore: * Stateless automation * Stateful orchestration * Long-running workflows * Process persistence * Workflow recovery * Correlation and context management An employee onboarding process may last days or weeks.A chatbot conversation may last minutes.These are fundamentally different workloads. WHY COPILOTS NEED A NERVOUS SYSTEM Human brains don't directly control every muscle individually.The nervous system coordinates actions.Enterprise AI requires the same model.This episode introduces the Logic App Nervous System architecture where: * Agents reason * Logic Apps orchestrate * Connectors execute * Policies govern * Identity secures * Observability monitors The result is coordinated intelligence instead of isolated automation. AZURE LOGIC APPS AS THE ORCHESTRATION LAYER Azure Logic Apps was originally designed for enterprise integration.It is rapidly becoming one of the most important foundations for agentic workflows.We examine: * HTTP-triggered orchestrations * Event-driven automation * Workflow persistence * Long-running process support * Enterprise connectors * Business process orchestration Logic Apps becomes the central coordination layer between agents and enterprise systems. STANDARD VS CONSUMPTION ot all Logic Apps are equal.Choosing the wrong hosting model can limit scalability before your architecture even launches.We compare: * Logic Apps Consumption * Logic Apps Standard * Stateful workflows * Stateless workflows * DevOps integration * Networking capabilities * Performance characteristics For serious agent orchestration, the answer becomes increasingly clear. STATEFUL WORKFLOWS: THE MEMORY LAYER Memory is what transforms automation into orchestration.Stateful workflows provide: * Checkpointing * Persistence * Recovery * Waiting states * Approval handling * Cross-system coordination We explain why workflow memory is often more important than model memory. THE AGENT LOOP ACTION One of Microsoft's most important innovations for agentic workflows is the Agent Loop action.This episode explores: * Think-Act-Learn cycles * Tool execution * Iterative reasoning * Memory retention * AI-assisted orchestration * Workflow-native agents Rather than bolting AI onto workflows, Agent Loop embeds reasoning directly into the orchestration layer. CONNECTORS AS NEURAL PATHWAY SIn the nervous system analogy, connectors become the nerves.They connect orchestration to execution.We discuss: * Microsoft Graph * SharePoint * Teams * Outlook * Dataverse * Dynamics 365 * Azure Services * Custom APIs The orchestrator becomes the central intelligence that routes activity across the enterprise. CUSTOM CONNECTORS AND LOGIC-IN-API Modern enterprises cannot expose proprietary business logic directly to agents.Instead, they need contracts.We explore: * OpenAPI specifications * Custom connectors * Internal APIs * Enterprise service layers * Reusable business capabilities * Governance boundaries Custom connectors become the contract layer between AI and enterprise systems. THE CROSS-TENANT CHALLENGE Most organizations no longer operate in a single Microsoft 365 tenant.Mergers, acquisitions, regional operations, and regulatory requirements have changed the landscape.This episode examines: * Multi-tenant architectures * Cross-tenant identity * Microsoft Entra collaboration * Sovereign boundaries * Tenant isolation * Enterprise coordination Cross-tenant orchestration is becoming the default, not the exception. MANAGED IDENTITIES EXPLAINED Secrets are one of the biggest weaknesses in enterprise automation.We explain how managed identities eliminate: * Client secrets * Credential sprawl * Manual rotation * Shared credentials * Configuration risk Identity becomes a platform capability instead of an operational burden. WORKLOAD IDENTITY FEDERATION Cross-tenant automation introduces a new challenge.How do workloads authenticate without secrets?This episode explores: * Workload identity federation * Azure AD Token Exchange * Federated credentials * Cross-tenant trust * Secretless authentication * Zero Trust architectures This becomes one of the most important building blocks for enterprise-scale agent ecosystems. MICROSOFT ENTRA AGENT ID Identity is becoming a first-class concern for AI agents.We examine how Microsoft Entra Agent ID enables: * Agent governance * Agent identities * Blueprint-driven permissions * Security boundaries * Authorization controls * AI accountability The future of AI governance begins with identity. ERROR HANDLING AS INTELLIGENCE Failures are inevitable.Resilience is optional.We explore advanced orchestration patterns including: * Scoped error handling * Adaptive retries * Compensating transactions * AI-assisted error triage * Self-healing workflows * Recovery orchestration The goal is not preventing failure.The goal is surviving failure intelligently. Become a supporter of this podcast: https://www.spreaker.com/podcast/m365-fm-modern-work-security-and-productivity-with-microsoft-365--6704921/support [https://www.spreaker.com/podcast/m365-fm-modern-work-security-and-productivity-with-microsoft-365--6704921/support?utm_source=rss&utm_medium=rss&utm_campaign=rss].

Building Multi-Agent AI Systems with Copilot Studio: From Ideas to Intelligent Automation with David Lorenzo Lopez [MVP]

Artificial Intelligence is rapidly evolving from simple chatbots into sophisticated multi-agent systems capable of automating complex business processes, collaborating across services, and delivering real business value. In this episode of the M365 Podcast, Mirko Peters sits down with Microsoft MVP David Lorenzo Lopez to explore the future of intelligent automation and how organizations can leverage Microsoft Copilot Studio, Azure AI Foundry, and the Microsoft Agent Framework to build scalable AI solutions.David shares his journey from web development and .NET programming to becoming a leading voice in AI-driven automation. He explains how the arrival of GPT models transformed the technology landscape and why the real challenge today is no longer generating impressive demos but creating measurable business outcomes with AI. WHAT ARE MULTI-AGENT AI SYSTEMS? One of the core topics of this conversation is the concept of multi-agent systems. David compares modern AI architectures to the evolution from monolithic applications to microservices. Instead of building one giant AI agent responsible for everything, organizations can create specialized agents focused on individual tasks and orchestrate them through a central coordinator.Key benefits include: * Improved scalability and maintainability * Better task specialization and accuracy * Easier testing and optimization * Reusable AI components across multiple business scenarios * Greater control over automation workflows COPILOT STUDIO VS AZURE AI FOUNDRY Microsoft now offers multiple ways to build AI-powered solutions, and David explains when to choose each platform.The discussion covers how Copilot Studio enables rapid low-code development using Power Platform integrations, while Azure AI Foundry provides greater flexibility, customization, and scalability for advanced AI implementations. As Microsoft continues to integrate these platforms, organizations have more options than ever to match their technical and business requirements.Topics covered include: * Copilot Studio connected agents * Azure AI Foundry orchestration * MCP connectors * Knowledge integration * Low-code versus pro-code development * AI workflow design patterns HUMAN-IN-THE-LOOP AND RESPONSIBLE AI While autonomous AI systems are becoming more capable, David strongly advocates for maintaining human oversight in critical business processes. He explains why AI should support decision-making rather than completely replace it, especially when financial, legal, or operational risks are involved.The conversation explores: * Approval workflows * Human validation processes * Governance strategies * Compliance considerations * Risk mitigation for AI automation MICROSOFT AGENT FRAMEWORK AND THE FUTURE OF AI DEVELOPMENT A major highlight of the episode is Microsoft's new Agent Framework. David explains how the framework combines capabilities from Semantic Kernel and other Microsoft AI initiatives to create a powerful platform for building enterprise-grade agents.Listeners will learn how developers can: * Create custom AI agents * Build complex orchestration workflows * Deploy scalable AI solutions * Integrate with Azure services * Develop reusable intelligent systems GOVERNANCE, SECURITY, AND THE EU AI ACT As AI adoption accelerates across Europe, governance and compliance have become essential topics. David discusses how Microsoft addresses security, data residency, privacy, and regulatory requirements through Azure AI services and emerging governance tools such as Agent 365 Control Plane.The discussion also covers: * Data protection requirements * European AI regulations * Azure OpenAI compliance * Model selection strategies * AI governance best practices CONTROLLING AI COSTS AND FINOPS One of the biggest challenges organizations face is understanding and controlling AI costs. David explains why estimating AI consumption is difficult and how businesses can establish practical monitoring and optimization strategies. Learn about: * Token consumption * Copilot Studio credits * Pay-as-you-go models * Cost optimization techniques * AI FinOps best practices KEY TAKEAWAYS This episode delivers practical insights for architects, developers, IT leaders, and business decision-makers looking to move beyond AI hype and create sustainable business value through intelligent automation.David's final message is simple yet powerful: AI is a wave that is transforming every industry. Organizations and individuals can either let it pass over them or learn how to ride it. Those who embrace AI responsibly, strategically, and thoughtfully will be best positioned for the future.CONNECT WITH M365 FMIf you enjoyed this episode, subscribe to M365 FM on Apple Podcasts, Spotify, YouTube, and your favorite podcast platform. Don't forget to leave a review and share the episode with colleagues interested in Microsoft Copilot, AI Agents, Azure AI Foundry, and the future of intelligent automation. Become a supporter of this podcast: https://www.spreaker.com/podcast/m365-fm-modern-work-security-and-productivity-with-microsoft-365--6704921/support [https://www.spreaker.com/podcast/m365-fm-modern-work-security-and-productivity-with-microsoft-365--6704921/support?utm_source=rss&utm_medium=rss&utm_campaign=rss].

The Rise of Private LoRA: Architecting Secure AI on Proprietary Data

Everyone is talking about AI adoption. Far fewer are talking about AI sovereignty. Organizations have rushed to deploy Microsoft Copilot, Azure OpenAI, ChatGPT Enterprise, Claude, Gemini, and dozens of AI-powered productivity tools. The results have been impressive. Productivity has increased. Development cycles have accelerated. Knowledge discovery has improved. But beneath the excitement lies a growing concern. What happens when your organization's most valuable asset—its proprietary knowledge—starts flowing into AI systems you don't fully control? In this episode, we explore the rise of Private LoRA (Low-Rank Adaptation), why data sovereignty is rapidly becoming one of the most important architectural challenges in enterprise AI, and how organizations can build secure, domain-specific AI models without training foundation models from scratch. We examine the convergence of AI governance, regulatory compliance, Microsoft cloud architecture, sovereign AI, LoRA fine-tuning, quantization, federated learning, and enterprise security. If your organization views proprietary data as a strategic advantage, this episode explains why the future of AI may not belong to the biggest models—but to the most specialized ones. THE SHADOW AI CRISIS Most organizations believe their AI strategy is governed. The reality is very different. Employees routinely paste sensitive information into public AI systems because they are faster and easier than approved tools. This phenomenon has a name: Shadow AI. We explore how: * Proprietary business data leaks into public models * Internal documents are shared outside governance boundaries * Competitive intelligence leaves the organization * Customer information becomes exposed * Security teams lose visibility The risk isn't always a breach. Sometimes it's simply the slow erosion of proprietary knowledge. WHY DATA SOVEREIGNTY MATTERS The conversation around AI is shifting. Organizations are no longer asking: "Can we use AI?" They're asking: "Where does the data go?" This episode explores the growing importance of: * AI Sovereignty * Data Residency * Data Localization * Cross-Border Data Restrictions * Intellectual Property Protection * AI Governance * Digital Sovereignty As regulatory pressure increases, organizations are discovering that data location is becoming as important as model performance. THE REGULATORY WALL IS ARRIVING Compliance is no longer a future problem. It's becoming an architectural requirement. We examine the impact of: * EU AI Act * GDPR * CPRA * LGPD * Data Localization Requirements * Financial Regulations * Healthcare Compliance Frameworks You'll learn why AI architectures designed for unrestricted global data movement may struggle in a world increasingly defined by jurisdictional boundaries. MICROSOFT'S APPROACH TO AI SECURITY Microsoft provides some of the strongest enterprise AI protections available today. But even with: * Microsoft 365 Copilot * Azure OpenAI * Azure AI Foundry * Microsoft Purview * Microsoft Entra ID * Azure Confidential Computing There remains a gap between approved enterprise AI usage and actual user behavior. We discuss how organizations can extend Microsoft's security model while maintaining control over proprietary intelligence. THE FALSE CHOICE BETWEEN PUBLIC AI AND BUILDING YOUR OWN MODEL Many organizations believe they have only two options: Option One Use public AI services. Option Two Build and train a foundation model from scratch. In reality, there is a third option. Private LoRA. This episode explains how LoRA enables organizations to customize powerful open-weight models without the extraordinary cost and complexity of full model training.  HOW LORA ACTUALLY WORKS  LoRA, or Low-Rank Adaptation, changes the economics of AI customization. Instead of retraining billions of parameters, LoRA introduces lightweight trainable layers that adapt an existing model to a specific domain. We break down: * Full Fine-Tuning * Parameter-Efficient Fine-Tuning * Adapter Architectures * Rank Selection * Training Efficiency * Model Specialization * Domain Adaptation The result is a highly customized AI model with a fraction of the cost and infrastructure requirements. QUANTIZATION CHANGES EVERYTHING LoRA becomes even more powerful when paired with quantization. Using techniques such as: * 8-bit Quantization * 4-bit Quantization * NF4 * QLoRA Organizations can dramatically reduce hardware requirements while maintaining strong performance. We explain how: * Memory consumption drops * Training costs decrease * Inference becomes affordable * Single-GPU deployments become practical This is one of the key innovations making sovereign AI achievable for mainstream enterprises. THE SINGLE-GPU ENTERPRISE AI MODEL  One of the most surprising insights in this episode is how little infrastructure is required. Using modern open-weight models and LoRA adaptation, organizations can: * Train on a single GPU * Deploy internally * Retain data sovereignty * Eliminate API dependencies * Reduce operating costs We explore architectures built around: * Llama * Mistral * Open-Weight Models * Azure GPU Infrastructure * Azure Kubernetes Service * Azure Machine Learning The economics are far more accessible than many organizations assume. Become a supporter of this podcast: https://www.spreaker.com/podcast/m365-fm-modern-work-security-and-productivity-with-microsoft-365--6704921/support [https://www.spreaker.com/podcast/m365-fm-modern-work-security-and-productivity-with-microsoft-365--6704921/support?utm_source=rss&utm_medium=rss&utm_campaign=rss].

The Death of the Dropdown: Why Manual Tagging is Killing Your Governance

or years, organizations believed metadata governance was a training problem.If users understood the taxonomy better, governance would improve.If the dropdown lists were clearer, metadata quality would improve.If more communication and documentation were provided, compliance would improve.But what if the problem was never the user?What if the real problem is that governance logic was placed in the wrong layer of the architecture entirely?In this episode, we explore why manual metadata tagging has become one of the biggest obstacles to modern governance, compliance, enterprise search, and AI readiness. We examine the collapse of traditional metadata models, the rise of Graph-powered governance, and how organizations are replacing manual tagging with automated classification, contextual intelligence, and real-time metadata injection.If your governance strategy still depends on users selecting values from dropdown menus, this episode may fundamentally change how you think about Microsoft 365 governance. THE MANUAL METADATA CRISIS Modern work has changed.Governance models haven't.Content is now created continuously across Teams, SharePoint, OneDrive, Outlook, mobile devices, and third-party integrations. Files arrive at a pace that no human-driven classification model can realistically keep up with.Yet many organizations still rely on users to manually classify: * Department * Project * Content Type * Sensitivity * Retention Category The result is predictable.Users skip fields.Users select defaults.Users guess.And governance slowly collapses under the weight of incomplete metadata.We explore why manual tagging doesn't fail because users are careless.It fails because the architecture assumes human behavior can scale indefinitely. THE HIDDEN COST OF DARK DATA Every untagged file creates a governance blind spot.The organization continues paying for: * Storage * Security * Backup * eDiscovery * Compliance Monitoring But receives none of the governance value metadata was supposed to provide.This episode examines the concept of dark data and how millions of documents become effectively invisible despite remaining stored and protected.Learn how missing metadata impacts: * Search * Compliance * Records Management * Retention * Analytics * AI Readiness And why many organizations are sitting on enormous repositories of information they can no longer govern effectively. WHY DROPDOWNS ARE A DESIGN FAILURE Most governance teams blame users.User experience research tells a different story.Dropdowns were designed to enforce consistency.Instead, they introduce friction.We discuss: * Decision fatigue * Metadata abandonment * Long taxonomy lists * User behavior patterns * Classification inconsistency * Cognitive overload The problem isn't that people refuse to govern content.The problem is that governance interrupts the flow of work.Every additional field creates another opportunity for bad metadata. THE COMPLIANCE IMPACT OF BAD TAGGING Poor metadata quality isn't just inconvenient.It creates regulatory risk.This episode explores how inconsistent classification directly affects: * Microsoft Purview * Data Loss Prevention (DLP) * Retention Policies * eDiscovery * Records Management * GDPR Compliance * HIPAA Controls When metadata is wrong, governance policies become unreliable.Sensitive data may be missed.Retention schedules may fail.Search results become incomplete.And compliance teams lose visibility into critical information assets. MICROSOFT GRAPH AS THE ORGANIZATIONAL NERVOUS SYSTEM Most organizations think Microsoft Graph is simply an API.In reality, it is a live representation of how work happens inside the enterprise.Graph understands: * Users * Teams * Groups * Files * Projects * Relationships * Permissions * Collaboration Patterns Instead of asking users to describe content, Graph can infer context automatically.We explore how Graph provides the foundation for a completely different governance model where metadata is generated from organizational signals rather than manual input. CONTEXT-AWARE GOVERNANCE Traditional metadata is static.Context is dynamic.A file's meaning depends on: * Who created it * Where it was created * Which project it belongs to * Who can access it * How it is being used This episode explains how governance systems can derive metadata automatically using Graph relationships rather than relying on user declarations.The result is richer, more accurate metadata that evolves as content moves through its lifecycle. AI-POWERED CLASSIFICATION Manual tagging isn't the only alternative.Modern AI services can classify content automatically.We explore: * Microsoft Syntex * AI Builder * Machine Learning Classification * Natural Language Processing * Document Understanding * Pattern Recognition * Sensitive Information Detection Learn how AI-driven classification improves consistency, reduces cost, and scales across millions of files. ARCHITECTING THE MIDDLEWARE LAYER One of the most important concepts discussed in this episode is the governance middleware layer.Think of it as a customs checkpoint for content.Before files are stored, middleware: * Intercepts uploads * Queries Microsoft Graph * Applies classification logic * Injects metadata * Assigns labels * Triggers governance policies All without requiring user interaction.We break down how Azure Functions, Microsoft Graph, webhooks, and event-driven architectures combine to make this possible. AZURE FUNCTIONS AND EVENT-DRIVEN GOVERNANCE Modern governance should happen at the moment content is created.Not months later during an audit.This episode explains how organizations are using: * Azure Functions * Microsoft Graph SDK * Webhooks * Delta Queries * Event Grid * Managed Identity To build real-time governance platforms that classify and enrich content automatically.The user saves the file.The platform handles governance. DYNAMIC PROPERTY INJECTION Metadata doesn't need to be manually entered.It can be generated.We explore how middleware automatically injects: * Project Codes * Department Ownership * Content Categories * Sensitivity Levels * Retention Schedules * Governance Attributes Using: * Property Bags * Schema Extensions * Open Extensions * Graph Metadata This creates a living metadata layer that remains accurate as content evolves. GOVERNANCE AT THE POINT OF ACTION Traditional governance is reactive.Modern governance is preventative.Rather than discovering problems months later, governance occurs at the exact moment content is created, modified, or shared.We discuss: * Real-time classification * Immediate policy enforcement * Automated retention assignment * Continuous metadata enrichment * Event-driven governance This shift fundamentally changes the economics of compliance and information management. SEARCH THAT ACTUALLY WORKS Most enterprise search failures are metadata failures.Search engines can only work with the information they receive.When metadata is incomplete, search becomes unreliable.This episode examines how automated metadata dramatically improves: * Microsoft Search * SharePoint Search * Knowledge Discovery * Content Discovery * Enterprise Findability * Information Retrieval The difference between searchable content and invisible content is often metadata. AI READINESS STARTS WITH GOVERNANCE One of the most important messages in this episode is simple:AI readiness is metadata readiness.Microsoft Copilot, AI agents, and retrieval systems depend on accurate content classification.Without metadata: * AI hallucinates more often * Search quality declines * Context is lost * Knowledge becomes fragmented With metadata: * AI retrieves better information * Recommendations improve * Summaries become more accurate * Organizational knowledge becomes accessible The future of enterprise AI depends on the quality of the governance layer beneath it. BUILDING YOUR AUTOMATION ROADMAP Moving beyond manual tagging requires a phased strategy.We walk through a practical implementation roadmap:Phase 1: AuditUnderstand your metadata gaps.Phase 2: Taxonomy DesignDefine the minimum metadata that drives governance.Phase 3: PilotAutomate one content type and one team.Phase 4: ScaleExpand automation across Microsoft 365.Phase 5: OptimizeImprove models, classifications, and governance policies over time.The goal isn't eliminating governance.The goal is removing governance from the user experience. Become a supporter of this podcast: https://www.spreaker.com/podcast/m365-fm-modern-work-security-and-productivity-with-microsoft-365--6704921/support [https://www.spreaker.com/podcast/m365-fm-modern-work-security-and-productivity-with-microsoft-365--6704921/support?utm_source=rss&utm_medium=rss&utm_campaign=rss].