76 sessions you won't want to miss

Keynote #PPFS2026

Opening Keynote at the Power Platform French Summit 2026.

Copilot Studio: How Did KESTREL Move from Initial Testing to a Structured AI Approach?

Everyone wants to experiment with generative AI. But how do you move from initial ideas to a structured, governed approach that’s truly useful for business units? Drawing on KESTREL’s experience, this session will provide a concrete account of the step-by-step development of an AI strategy centered on Copilot Studio: identifying initial use cases, conducting experiments, supporting teams, and establishing a governance framework that enables continued innovation while keeping it under control. Presented jointly by KESTREL and MOCA, the session will combine the client’s perspective with that of its partner to provide a candid look back at the journey so far: the choices that were made, the challenges encountered, the experiments that delivered value, as well as the pitfalls, adjustments, and lessons learned from real-world experience. This is a decidedly pragmatic session designed to share not a theoretical vision of AI adoption, but what actually happens when a company begins to put Copilot Studio in the hands of its business units.

How to Organize a Successful Agentathon

Giving all your makers access to Copilot Studio and Agent Builder is good. Giving them the motivation, the framework, and the momentum to create useful agents is even better. The Agentathon is the ideal accelerator: in just a few days, it brings concrete use cases to light, uncovers your future champions, and generates a level of excitement that ten training sessions couldn’t achieve. During this session, you’ll receive comprehensive feedback from the field and a complete guide covering everything from format and preparation to licenses, environments, facilitation, and the judging panel—and, most importantly, what comes next—so that the momentum carries on beyond the final pitch.

Three Sectors, Three Copilot Studio Projects: What Worked, What Went Wrong, What Was Missing

An industrial organization, a service provider, and an IT department. Three Copilot Studio projects. Three different trajectories. One has spawned others. One never made it past the POC stage. The third is about to go live after several twists and turns. Through these three experiences, we’ll take a step back to examine the factors that truly tipped the scales: the common elements found across projects, the specific business requirements, and the obstacles that sometimes arise where we least expect them.

REX Vinci Construction - Eliminating the technical debt of a critical business application using Power Apps Code Apps

How can you ensure the reliability of an environmental tool while preparing for future regulatory and business changes? Vinci Construction shares its insights on the overhaul of a strategic Power Platform application.

Contract 15: When Contract Data Becomes a Decision-Making Tool

How can you turn the tracking of complex contracts into a true business management tool? In the construction industry, tracking subcontracts and service agreements involves managing a large amount of information, financial commitments, and expenses throughout their execution. When this data is scattered across various systems, it can be difficult to maintain a reliable and up-to-date overview needed for effective management. To address this challenge, Vinci Construction and Uni5 developed Contract 15, a business solution that centralizes key contract information and provides a comprehensive view of contract progress. With a real-time view of contracts, committed amounts, and actual expenditures, Contract 15 provides managers with reliable metrics to ensure effective contract monitoring, control costs, and facilitate decision-making. Through this case study, discover how a digital solution designed around business needs can transform contract data management into a powerful management tool.

REX CodeApps: The Bet That Transformed Our Apps

Why did we choose to rewrite our applications using CodeApps? In this case study, we share the tangible benefits we’ve seen for our users and our organization: faster applications, a redesigned user experience, and the ability to evolve our tools more easily over time. A jargon-free overview to help you understand what such a transformation can truly bring to your business.

Building a Copilot Business Case: Methodology, Measurable Benefits, and Cost Model for Scaling Up

Organizations want to adopt Copilot to boost productivity, automate low-value tasks, and improve the quality of decisions. But one question comes up time and again: How can they justify the investment? This session offers a practical methodology for building a robust Copilot business case based on data, operational scenarios, and a clear financial model. We’ll explore how to identify the right use cases, quantify the benefits (time, quality, revenue, risk reduction), and model the costs (licenses, integration, governance, adoption). Participants will leave with a replicable approach to convince their leadership, secure budgets, and drive value over the long term.

From Contact Center to Autonomous Service

AI is no longer just assisting agents—it is becoming the agent. In this session, we’ll explore how Dynamics 365 Contact Center is evolving into a truly agent-driven platform, where AI agents can understand intent, orchestrate conversations, take actions, and continuously optimize customer journeys. You’ll discover: - How AI agents (CIA, KMA, CMA…) are transforming customer service operations - How orchestration, real-time signals, and Copilot enable end-to-end automation - Real-world strategies for effectively combining human and AI agents - What this means for architecture, governance, and future contact center design If you think contact centers are just about routing and queues, think again :-) This is the shift toward autonomous customer engagement.

Business AI: Beyond the Hype, What Are the Concrete Benefits for Finance, Supply Chain, Sales, and Operations Departments?

AI is everywhere. But there’s sometimes a big gap between spectacular demonstrations and the reality on the ground. Where can real value be created? Which processes should be automated? What benefits can we expect? And how can we avoid projects that don’t deliver a return on investment? During this roundtable discussion, our experts will share concrete examples of how AI is being used to support business operations. We’ll cover the following topics in particular: > Copilot Studio: custom-built agents for business units, with insights from several real-world examples. > “Off-the-shelf” AI agents developed by Microsoft in D365 > Challenges related to governance, security, and cost control; > The conditions for success in moving from experimentation to large-scale adoption. A business-focused discussion session designed to help decision-makers distinguish between passing trends and opportunities that truly create value.

ROI and Productivity: The Vibe Coding Approach for Your Power Pages Portals

Discover how to maximize the return on investment for your web portal projects while boosting your teams’ productivity. This session presents Vibe Coding’s innovative approach to Power Pages, transforming the way your organizations design and deploy their digital solutions. Key topics covered: - Reduced development costs - How to cut the time it takes to build your customer, partner, or employee portals by up to 70% - Faster time-to-market - Moving from concept to production in weeks rather than months - Resource optimization - Enabling business teams to contribute directly without relying on IT every step of the way - Simplified governance - Maintaining control and security while promoting team autonomy

Data, AI, and Real-World Challenges: Why Our HR Chatbot Failed

In 2025, I launched an ambitious project: to develop an HR chatbot based on Copilot Studio to automatically answer employees’ questions and reduce the workload on HR teams. From a technical standpoint, the project checked all the boxes: Microsoft 365 integration, Power Platform architecture, RAG engine, and access via SharePoint and a web portal. However, the expected success never really materialized. In this session, I’ll share an honest account of the challenges we faced: outdated documentation, scattered information, a lack of document governance, insufficient business buy-in, and sometimes unrealistic expectations surrounding generative AI. Through this real-world case study, we’ll explore why data quality remains the key factor in the success of internal chatbot projects—and why Copilot Studio cannot compensate for a flawed document base. Participants will leave with a practical assessment framework to evaluate their organization’s readiness before launching a chatbot or AI agent project.

IT Leaves the IT Department: Skills, Staff, and New Makers

With generative AI, bots, and skills, what used to be called “Shadow IT” now seems quite tame. Business units are no longer content to simply use the tools provided by the IT department—they’re building their own solutions, automating their processes, and creating their own apps and bots. In short, welcome to the era of the new makers. This session explores this new decentralization of IT and the issues that come with it.

One idea, six ways to bring it to life: How do you make the right choice?

Five years ago, designing a Power Platform solution often came down to choosing between a Canvas App and a Model-Driven App. Today, the same business need can be addressed in multiple ways: Canvas or Model-Driven apps, AI-assisted generation, conversational agents, pro-code integration, or the creation of on-demand tools. The challenge is no longer about what can be built, but about determining which approach best fits the project’s context, constraints, and ambitions. Through concrete examples and real-world insights, this session explores the different ways to design Power Platform solutions in the age of AI: speed of implementation, governance, maintainability, user experience, scalability, and the developer’s role. The goal is not to identify a “winning” approach, but to provide guidelines for making the right choices at the right time. By 2027, the issue will no longer be about choosing a technology; it will be about choosing the right approach for the right need.

Can AI Help Clear 40 Years of Technical Debt? Insights from Fedd’s CIO

40 years of code, a massive 4D legacy, and almost no documentation: at Fedd (AGON Group), technical debt had become a major challenge. To address this, Fedd took a chance on AI with Uni5 Technologies, designing an agent capable of analyzing the code, generating documentation for it, and keeping it up to date with every commit. During this session, Fedd’s CIO will speak firsthand about this decision, the challenges encountered, and the initial results achieved. A real-world case study on AI applied to a 40-year-old legacy system that is still active and mission-critical.

Copilot Studio "V2" Takes It to the Next Level: Agents, Workflows, Skills… and Apps

Copilot Studio is no longer limited to creating conversational agents. With the arrival of the GitHub Copilot harness and new capabilities related to skills, tools, workflows, memory, and integration with the Microsoft ecosystem, the platform is undergoing a profound transformation. But in practical terms, what can we build with it today? Through a session deliberately focused on demonstrations and real-world use cases, we’ll explore the key new features of Copilot Studio: agents capable of reasoning through complex tasks, reusable skills, orchestration with Agent Flows and new workflows, file manipulation and generation, integration of external tools and services, collaboration between agents, and the creation of application experiences that can be shared with colleagues. The goal: to show how these different building blocks can now be assembled to go far beyond a simple chatbot and build true AI solutions capable of reasoning, automating, producing, and taking action. We’ll conclude with a practical overview of new consumption models, anticipated costs, and the impacts of this evolution on the governance and industrialization of Copilot Studio solutions.

Cellnex × Power Platform: Three Years to Reshape the IT Landscape

For the past three years, Cellnex France has been building its information system on the Power Platform using an agile approach. First, Canvas Apps on Dataverse centralized business data, and a data well was used to share it across the company. Next, AI Builder automated document extraction. Finally, generative AI arrived, with Azure AI Foundry, AI Builder prompts, and now no-code development with Code Apps. Driven by a holistic vision, this approach has yielded transformative results that are now at the heart of the company’s processes. David Verbeke, Application Manager at Cellnex France, and DevPoint will present these results. They’ll also explain how the Power Platform is reshaping the IT landscape and what new opportunities it has unlocked—for both decision-makers and technical professionals.

Copilot Credits at Scale: Controlling Costs and Accelerating Adoption

With the rise of Copilot Studio, Copilot Cowork, and Work IQ, Copilot Credits consumption is becoming a key challenge for businesses. This session presents an AI FinOps approach to monitor, govern, and optimize this consumption at scale, particularly using tools from the Microsoft 365 Admin Center and the Power Platform Admin Center. Learn how to anticipate needs, allocate capacity, and empower teams while accelerating the adoption of Copilot Studio agents, apps, and workflows, as well as Copilot Cowork and Work IQ—without losing control of costs.

The Illusion of Productivity with AI: Measuring the True Value of an Automated Process

Automation can save a person ten minutes without speeding up the process for the customer. Using a business workflow built with Power Apps, Power Automate, and Copilot Studio, this session distinguishes between local gains and end-to-end value. Which timelines should be measured? Where do validations and rework occur? How can we compare quality, cost, and user experience before and after adding an agent? A concrete metrics framework will help decision-makers prioritize use cases that produce observable results.

[REX]: Copilot Studio in Real-World Conditions: What We've Really Learned from Our Client Projects

There’s sometimes a world of difference between demos and an agent actually being used in a business setting. In this session, we’ll share several insights from Copilot Studio projects conducted with clients, covering a wide range of use cases: document research, analysis of complex documents, business assistants, process automation, and even the deployment of agents on a larger scale. Through concrete examples, we’ll explore what actually works, the architectural choices that had to evolve during the project, the limitations encountered in the field, and the solutions implemented to overcome them. We’ll also address the issues that quickly arise once you move beyond the POC: knowledge quality, orchestration, costs, security, governance, maintainability, and adoption. This session is based on real-world experience—including both successes and challenges—to provide a deeper understanding of what it truly means to industrialize a Copilot Studio agent.

The 10 Commandments for Success as a Copilot Studio Agent

In the beginning, there was a POC. It looked great, performed well in demos, and everyone loved it. Then came the day it went live… and the agent didn’t meet users’ expectations. If this story sounds familiar, this session is for you. Drawing on real-world projects carried out for clients of all sizes, I’ll walk you through the fundamentals for making your chatbot project a success. Spoiler: there are more than just 10. Mistakes to avoid, best practices, and a wealth of real-world insights. What works—and what doesn’t. Everything you need to know to ensure your chatbot doesn’t remain just a proof of concept. Walk away with a checklist to make your chatbot a success every time! Feel free to share your feedback and experiences—both positive and negative.

REX Getlink: Power Apps Application: When Plan B Becomes the Permanent Solution.

A tool for managing end-of-life capabilities, a non-negotiable shutdown date, and a simple rule: without monitoring employee capabilities, operations stop. No degradation of service, no manual fallback mode. A complete shutdown. The target solution would not be ready in time. So we set up a parallel Plan B on Power Platform—intended as a temporary measure—with a single goal: to ensure continuity on the day of the switchover. D-Day went smoothly. What we hadn’t anticipated was what happened next: users embraced the transitional solution and refused to switch back. Plan B became Plan A. In this session, we’ll walk through the entire process: how to define the scope when failure isn’t an option, the architectural trade-offs made under tight deadlines, what we deliberately left out, and the moment when we had to transform a fallback solution into a sustainable product. An unfiltered look back at our experience, including what we’d do differently and the questions that remain unanswered: What if our mistake was treating the Power Platform as a Plan B? What if the constraints ultimately led us to a better solution?

Licensing & AI: How Much Does an Agent Really Cost? GitHub Copilot, Copilot Studio, and the Shift from Per-User to Per-Use Billing

In two years, AI has completely upended the Power Platform licensing model. We’ve gone from a simple, predictable model—one license, one user, one app—to a world where cost is based on usage: messages and Copilot Credits in Copilot Studio, AI Builder capacity, autonomous agents, and pay-as-you-go. In investment committee meetings, the question is no longer “How much does the license cost?” but “How much will this agent cost me once it’s in production, at scale?” In this session, I’ll break down the three trends reshaping the billing landscape: 1) The GitHub Copilot harness and development agents: who pays for what when the code, the application, and the agent are generated? Where does the developer seat end, and where does platform consumption begin? 2) App building in Copilot Studio: what’s included, what’s consumed, and the common pitfalls—autonomous agents, premium actions and connectors, Dataverse, and test environments that consume resources just like production. 3) The shift in Power Platform licensing toward pay-as-you-go: how to estimate, cap, and manage variable consumption without hindering adoption. On the agenda: a clear overview of the licensing models (per user / per capacity / per message / pay-as-you-go), concrete order-of-magnitude figures from real-world scenarios, a reusable TCO estimation model, and governance levers (Managed Environments, capacity quotas, consumption tracking, environment strategy) to avoid unpleasant surprises at the end of the month. You’ll walk away with three practical skills: choosing the right licensing model for your scenario, calculating the cost of an agent BEFORE building it, and explaining the invoice to your finance department.

Agent Builder or Copilot Studio? Let's stop choosing at random

Creating an agent has never been easier… but which tool should you choose? Agent Builder or Copilot Studio? Drawing on several real-world examples we’ve encountered—ranging from a simple information assistant to an agent connected to the company’s IT system—we’ll share our experiences, the choices we made, the limitations we faced, and the lessons we’ve learned from our projects. The goal: to understand when a simple approach is sufficient, when Copilot Studio becomes necessary, and how to successfully transition from prototype to production.

From Meetings to Insights: How AI Enriches Customer Data and Guides Action

How can we go beyond a simple meeting summary? This session demonstrates how AI transforms a meeting into insights, tasks, business signals, and recommendations—all enriched with Dynamics 365 context to provide concrete guidance for next steps.

Microsoft Fabric: The New Enterprise Data Asset Manager?

For a long time, data platforms were primarily designed to collect, transform, and analyze business data. Microsoft Fabric is part of this history, but its scope is evolving, prompting us today to view the platform in a new light. What if we stopped thinking of data as mere technical resources and began to see it as a true corporate asset? An asset that must be understood, organized, made accessible, secured, enriched, leveraged, and maximized. Centered around OneLake, Microsoft Fabric already brings together the worlds of data ingestion, data engineering, the Lakehouse, the data warehouse, data science, real-time analytics, and Power BI. But this ecosystem is also expanding into operational data and AI. In this session, we’ll deliberately take a step back from the specific features to ask a more strategic question: Is Fabric becoming the steward of the company’s data assets? We’ll explore how the convergence around OneLake—including the catalog, governance, semantics, and various workloads—is gradually bringing analytical and operational data closer together. Finally, we’ll explore why this vision takes on a new dimension with AI and agents, which are no longer content to simply display data but must also be able to understand it in context and leverage it.

Agentic Engineering: Building an AI-Enhanced Digital Factory at Safran

What if AI were no longer used solely to generate code, but to enhance the entire delivery cycle—from the expression of a business need through development, testing, and documentation? Drawing on Safran’s experience, this session will present a vision of a Digital Factory orchestrated by specialized AI agents. Analyst, Product Owner, architect, UX/UI designer, developer, tester, or technical writer: each agent handles a specific task, while human teams retain responsibility for validation and decision-making. We’ll demonstrate how a business request can be qualified, structured, and then gradually transformed into a product brief, specifications, a backlog, functional and technical rules, tests, and documentation. The session will also address the conditions necessary to scale this approach.

Managing 500 AI agents before it's too late

AI agents are proliferating at an unprecedented rate. With Microsoft 365 Copilot, Copilot Studio, and the Power Platform, anyone in any line of work can now create and deploy an agent in just a few minutes. The challenge is no longer building an agent. Drawing on feedback from various case studies in manufacturing, insurance, and enterprise environments, we’ll share the patterns we consistently observe in the field—such as an explosion in the number of agents, shadow IT, difficulty controlling costs, and lax governance—as well as our general recommendations to clients on how to regain control without stifling innovation.

Which Agent for Which Problem? A Decision Guide to Avoid Building One Too Many Agents

The session presents a decision tree to help choose between Copilot, a declarative agent, an autonomous agent, multi-agent orchestration, and a conventional workflow. The starting point is this: the question is no longer simply whether we can build an agent, but whether we should do so and which category is appropriate. Two common causes of disappointing projects are identified: assigning a task to an autonomous agent when a deterministic workflow would have sufficed, and building a multi-agent orchestration when a well-designed instruction would have been sufficient. The proposed decision-making framework is based on four dimensions: task ambiguity, the cost of an error, auditability, and execution frequency. The session compares several families of solutions (conversational assistants, declarative agents, tool-using agents, long-running autonomous agents, multi-agent orchestration, and traditional workflows) and identifies, for each, the relevant use cases and situations where the complexity becomes difficult to justify. The final section addresses keeping a human in the loop, measuring the quality of an agent in production, and cases where the right decision is not to build an agent. Format: 45 minutes, scoping session with feedback from the field. Level: all levels.

From Process to Agent: Do We Really Need to Turn Everything into an Agent?

Agent-based AI is everywhere, and it’s very tempting to turn every process into an agent. However, a good old Power Automate flow is often more reliable, less expensive, and easier to maintain. In this session, Wakers uses examples to show when an agent adds real value and when it introduces unnecessary complexity. We’ll also explore how to combine flows and agents in a hybrid model, and how converting a flow into an agent flow affects billing. You’ll leave with simple criteria to help you make the right choice.

Demystifying the Electronic Invoicing Reform: Keys to Accelerated Implementation with the Power Platform

From Confusion to Mastery: Behind the Scenes of a Multi-ERP Integration. A Look Back at the Electronic Invoicing Project: Implementing an Integration Solution Between 4 ERP Systems and the Authorized Electronic Invoicing Platform. Context, Architecture, Success Factors, Challenges, Conclusions. Why this session? The ease of implementation with the Power Platform allowed us to focus on the specific requirements of the client’s information system, its sales administration, and the 200 pages of the AFNOR standard (and its 92 appendices), making it possible to meet the reform’s implementation deadlines.

REX - Optimizing the Medical Care Process for Flight Crew Members at CEMA (Air France)

Find out how Air France’s Aviation Medical Expertise Center leveraged the Power Platform to design and implement a solution for real-time optimization of the medical care process for its flight crew (over 7,000 people per year).

From POC to Value: A Field Report on the Scaling Up of Copilot Studio Agents

Creating an agent is simple. Getting it adopted and generating real value is another matter entirely. Drawing on concrete real-world feedback, this session explores the transition from proof of concept (POC) to production using Copilot Studio: challenges encountered, mistakes made, user adoption, governance, and measuring value. An unfiltered perspective to understand what really works in a business setting.

Copilot Agents: Stop winging it—start by setting the framework.

Inconsistent responses, security risks, poorly estimated costs, business unit disappointment… Creating agents in Copilot Studio seems quick and easy, but how many actually deliver on the promise made to business units? This session analyzes the mistakes that turn a promising project into an underutilized agent that’s costly to maintain or unable to meet business unit expectations. You’ll learn how to move from a vague idea to a structured project: real-world scenarios, identified causes of failure, and a directly applicable scoping method. You’ll leave with a step-by-step operational guide to lay a solid foundation for your agent projects before you even open Copilot Studio.

From CRM to AI Agents: Why Your Business Data Is the Key to Your Transformation

What if the data and tools you already use were the starting point for your AI transformation? Companies have been investing in their CRMs, applications, and business processes for years. With the advent of generative AI and intelligent agents, new opportunities are emerging to better leverage these investments, streamline operations, and improve team performance. But how can you turn these opportunities into tangible results? This session will demonstrate how Dynamics 365 and Power Platform can serve as a foundation for gradually integrating AI into the core of business processes—without starting from scratch. Through real-world case studies and lessons learned, we’ll discuss the benefits already observed, the challenges to anticipate, and the steps to take to move toward more advanced use cases—particularly AI agents—with a clear goal: to create business value and achieve measurable results.

Which CRM-ERP integration is best for which use cases? No-code, low-code, or pro-code?

Insights into the various ways to integrate a CRM and an ERP system, based on use cases and technical solutions. Functional demonstration. Agenda: - Business challenges of CRM-ERP integration (synchronization of customer and product databases, synchronization of sales processes) - Technical solutions for CRM-ERP integration (No-Code, Low-Code, Pro-Code) - Demonstration of a Low-Code solution (Power Automate flows between Dynamics 365 Sales CRM and Business Central ERP) for three use cases.

Beyond Technology: Putting People at the Heart of AI Adoption at L'Oréal

The adoption of AI is not solely about technology. Its success depends above all on an organization’s ability to engage its employees, evolve its practices, and create the conditions for sustainable innovation. Drawing on L’Oréal’s experience, this session will explain how the company has gradually built an AI adoption strategy by leveraging the Microsoft ecosystem and the LCNC practices that have been in place for several years. We will share the key factors that enabled the transition from experimentation to implementation: governance, security, data protection, community engagement, training, change management, sharing best practices, and reusing existing assets. The session will also highlight how AI is currently transforming design and development processes by augmenting teams rather than replacing expertise, and by leveraging years of investment in low-code, no-code solutions, and automation.

Copilot Cowork: Connect your data and business processes with plugins and skills

How can you connect Copilot Cowork to your company’s data, tools, and processes? Learn how to create custom plugins and skills to give Cowork access to your business systems, enrich its knowledge, and enable it to take direct action on your processes. Through hands-on demonstrations, we’ll explore how to design, secure, and deploy these extensions to tailor Cowork to your organization’s needs.

Rethinking the Integration of AI Projects: An Agent-Based Approach for the Center of Excellence

With the growing adoption of AI, Centers of Excellence dedicated to agents are in increasingly high demand. Employees often come in with an idea, without necessarily knowing whether an agent is the right solution, whether a solution already exists, what they can build on their own, or when they need support.

REX Fabric Apps: When Power BI Is No Longer Enough, a BI Team Switches to Fabric Apps

At Klépierre, Power BI is everywhere. But when the legal and internal control teams asked to visualize the group’s entire organizational chart and the cash flows between its entities—across multiple countries and according to various analysis criteria—the standard visuals reached their limits, and developing a custom visual would have been time-consuming and costly. So we took a different approach: a Fabric App connected directly to the existing semantic model, developed with the help of coding agents. The result: exactly the visualization the Legal Committee was looking for—and even more. In this case study, we’ll explore how a team with a history of using Power BI—and no web developers on staff—took the plunge: what coding agents make possible, how the semantic model remains the single source of truth, and how to decide between a Power BI report and a Fabric App.

From business requirements to a high-level cost estimate in just a few minutes—Copilot, the super assistant

Developing Copilot Studio agents like code: YAML, CLI, and AI coding tools in action

What if you could design, version, and deploy your Copilot Studio agents directly from your terminal, using GitHub Copilot CLI or Claude Code as a true development copilot? That’s exactly what the open-source project Skills for Copilot Studio (Microsoft) enables: a plugin for AI coding tools that exposes Copilot Studio as a code target—topics, actions, knowledge, triggers, and variables, all in YAML, with schema validation, templates, and specialized sub-agents. In this technical session, we’ll cover: - The plugin’s architecture and its four sub-agents (manage, author, test, advisor) - How to clone, edit, and push a Copilot Studio agent from the terminal using GitHub Copilot CLI, Claude Code, or VS Code - The complete workflow: YAML authoring → validation → push → automated tests - Best practices for treating an agent as a code project: versioning, CI/CD, code reviews - Live demos of real-world scenarios, from cloning to deployment A session for developers and architects who want to go beyond the graphical interface and industrialize the creation of AI agents.

My name is Agent, Agent 365: Licensed to govern

From his seat at the bar, martini in hand, Agent 365 watches every agent who walks through the door. Q made sure of that: a pair of glasses that see right through any disguise, giving him a clear view of what every agent is up to, no matter how they’re dressed. Agent 365 determines who gets to stay, decides what each agent is allowed to access, and quietly revokes the privileges of anyone who oversteps their bounds. And when something tries to slip past him—a forged credential, a hand reaching for data it shouldn’t touch—he’s already moving to secure the room, backed by Entra, Purview, and Defender. Moneypenny invited us for a sync at the Prime Minister’s office, but after that, we’ll jump right into action. We’ll walk through what Microsoft’s new control plane for AI agents actually does. No martinis required—just a clear picture of how to manage the growing population of agents in your organization. Shaken, not stirred.

From Handcrafted Canvas to the Code Apps Factory

The annual hackathon with our Citizen developers prompted us to make a quick pivot to integrate AI tools like Claude Code into our app development process. From an interest in code apps to the creation of a VS Code kit that enables us to build projects quickly—all hosted in our Power Platform environments. A kit for everyone: pro devs, product owners, and citizen devs

TheFork Case Study: The Snowflake x Power BI x Claude Trio

Charles Brault (Director of Data, TheFork), joined by Tristan Malherbe (CEO of Data Pulse), will demonstrate how his team at TheFork built a modern, powerful, and agile data platform using the Snowflake x Power BI x Claude trio. On the agenda: real-world business cases and demonstrations to show how to integrate AI and LLMs into data and best address business questions.

Designing Power Platform Solutions in 2027: Proven Patterns, Practices, and AI Integration

This highly practical session explores advanced architectural patterns for Power Platform solutions, including modular application design and layered architecture, using very concrete examples. This session highlights the synergy between Power Apps Vibe coding, Power Apps Code Apps, Power Automate, and Copilot Studio. We’ll also explore how AI is transforming solution design, sharing industry best practices with clear dos and don’ts to help you avoid common pitfalls. As AI becomes a core element of the low-code development lifecycle, a critical question arises: how far should we go—and what new challenges does AI introduce? To answer this, we rebuilt a typical application both with and without vibe coding and will share the lessons learned from this comparison. Finally, the session includes a live demonstration of a complex workflow automation powered by a state machine integrated with AI—originally created in 2025 and enhanced in 2026, delivering remarkable benefits. This example illustrates how advanced patterns and AI can work together to create scalable, intelligent solutions. Power Apps vibe coding, Power Apps code apps, Power Automate, and Copilot Studio will all come together in this session.

AI agents that never leave the workstation: REX Microsoft Foundry on Windows: SLM + NPU + custom framework

DevPoint and TD SYNNEX have jointly designed and tested a series of PC user assistance agents on machines provided by Microsoft and TD SYNNEX. These agents serve a variety of purposes but share a common approach: inference remains on the machine, data does not leave the system, and the cost per query is zero. We’ll share this unfiltered feedback: what the Microsoft Foundry on Windows framework (SLM and NPU) truly enables and where its limitations lie, demonstrations of agents in action, key technical considerations (model selection and quantization, NPU execution, orchestration, packaging, and deployment), and insights for decision-makers: which use cases justify on-premises processing, how this impacts sovereignty and costs, and what hardware infrastructure is required. Leave with a clear understanding of what’s feasible right now—and what’s best left to the cloud.

Trust but Verify: Testing Agents in Copilot Studio

Agents look impressive in demos. In production? That’s where things get interesting. An agent that usually works isn’t the same as one you can trust. When you put agents in front of real users, things change. Prompts behave differently, grounding gets creative, and that one action you were sure about suddenly isn’t reliable anymore. Agents can now use skills as well, which is one more thing to validate before you release. We’ll explore what it really takes to test in Copilot Studio—not just running a few ideal scenarios and hoping for the best. That means examining how agents behave when users don’t follow the script (and they won’t), and how to validate prompts, grounding, actions, skills, and orchestration against what actually happens. Copilot Studio has added a lot of new tools: * Agent Evaluation Sets let you run up to 100 test cases in batches. * The Live Orchestrator Activity Map shows why the LLM chose that knowledge source, skipped that topic, or formatted its response the way it did. * The Test Hub handles standalone prompts and plugins before they ever reach a topic. From “it seemed fine when I tested it” to something you can actually ship without holding your breath. 😉

From Data to Decisions: Predictive Forecasting with Microsoft Fabric, Dataverse, and Power Apps

This topic presents an end-to-end forecasting solution using Microsoft Fabric, Dataverse, and Power Apps. Business data stored in Dataverse can be connected to Fabric for data preparation, analytics, and predictive modeling. Forecast results can then be stored and made available to business applications, allowing users to visualize predictions and take action directly through Power Apps and automated workflows. The goal is to transform historical operational data into actionable forecasts that support better and faster business decisions. This architecture closely follows Microsoft’s reference pattern for predictive analysis using Dataverse, Fabric, predictive models, and Power Platform applications.

Chateauform's Experience with Power BI & Fabric

Chateauform (premium corporate seminars and experiences) has been using Power BI since 2017 and recently migrated its data stack to Microsoft Fabric. Maxime Ngate, BI Director at Chateauform, will explain how his team uses the Power BI and Fabric combination on a daily basis, detailing various use cases and the value created for the different business units.

Beyond the Canvas: Let AI Do the Heavy Lifting in Power Automate

Power Automate makes it easy to build your first flow. The real challenge lies in designing, maintaining, optimizing, and evolving complex automations. In this session, you’ll see how AI becomes your Power Automate co-builder throughout the entire lifecycle. Using a real-world approval process with Outlook Actionable Messages (Adaptive Cards), we’ll demonstrate how AI helps design solutions, write complex expressions, resolve platform challenges, and simplify maintenance and updates—so you can build more capable automations with confidence. Three key takeaways: Use AI throughout the entire Power Automate lifecycle—from design and development to maintenance, troubleshooting, optimization, and updates. Build advanced approval processes with Outlook Actionable Messages and Adaptive Cards. Learn a practical AI-assisted approach that makes complex Power Automate solutions easier to build, maintain, and evolve over time.

Agentic Engineering for Dataverse: Production-Ready Code Generated by AI

Stop crossing your fingers and hoping that AI-generated code will hold up in production. In this session, we’ll cross the line between “vibe coding” (random prompts, unpredictable results) and agentic engineering: a systematic approach where AI doesn’t just write code that works, but code you’d be proud to deliver to an enterprise client. For several months now, I’ve been deploying code written by AI agents to real Dataverse client environments. The secret? An engineering approach where architectural constraints, deployment, and testing are no longer after-the-fact safety nets. They become the very ground on which the agent operates. What you’ll see: A live demonstration of an AI agent writing production-ready Dataverse plugins and custom APIs: typed step logging, declarative images, integrated Dependency Injection (DI), and test coverage that most human developers never achieve. What you’ll learn: Why tools matter more than the model. What sets agentic engineering apart from ad hoc prompting. How to build this environment within your organization and make AI a reliable member of your Dataverse development team.

Managing the Proliferation of AI Agents—The IT Survival Guide

Agent Builder, Copilot Studio, SharePoint agents: In just a few months, any employee can create an AI agent. From an IT perspective, the question is no longer whether to govern, but how to do so without stifling innovation. In this session, we’ll share a practical framework for structuring your agent governance based on your goals and constraints: what technical levers are actually available, how to manage credit consumption, what licenses are required, and how to organize the agent lifecycle from request to decommissioning.

Master Power Pages with AI: Easily Create Websites That Make a Difference

Discover how to easily create websites that are visually appealing, modern, and functional using Microsoft Power Pages and artificial intelligence. This session will show you how to leverage AI-powered tools to speed up website creation, customize their design and layouts, and integrate dynamic features with minimal effort. We’ll explore how AI can help you build professional, intuitive web experiences tailored to your users—whether you’re new to Power Pages or an experienced developer. The goal: to use AI to work faster with Power Pages without sacrificing design, code quality, or customization.

The Art of Organizing Microsoft Fabric: Workspaces, Capacity Planning, Governance, and Data Mesh

Every Microsoft Fabric program eventually faces the same deceptively simple question: How should we organize our workspaces? The answer shapes collaboration, security, lifecycle management, ownership, capacity allocation, cost control, and team autonomy. This session provides a practical decision model for combining Fabric workspaces, capacities, and domains without confusing their responsibilities. Workspaces remain the primary boundary for collaboration and access. Capacities define multi-tenancy: which workloads share compute resources, how they are segregated, and how far an incident can spread. Domains support organization and delegated governance, but they are not a security boundary. Data Mesh adds product ownership and federation without requiring every domain to have its own capacity—or its own miniature platform. We will also bring capacity planning back into the conversation. SaaS changes how infrastructure is delivered; it does not eliminate the need to understand demand, contention, criticality, failure domains, and cost. Nor is planning solely about slow reports: poor segregation can turn a single noisy or failing workload into a platform-wide incident. Copilot, Data Agents, and MCP-based consumers introduce less predictable demand. Attendees will leave with a reusable framework for making explicit organizational choices instead of accumulating workspaces and hoping that naming conventions will somehow form an architecture.

Scaling Administration: Automation Using the New Power Platform Admin APIs

The Dataverse environment-level APIs have been around for as long as organizations and environments have existed. Now they are grouped within a tenant with tenant-level settings that apply across multiple environments, as visualized in the Power Platform Admin Center. In Power Automate, you can use the Power Platform Admin connector to manage all these settings, create new environments or modify existing ones, and perform many other admin-level actions previously reserved for Global Admins. Why spend 6 minutes doing something when you can spend 6 hours automating it? Join Yannick in this session for a look at the API story behind it all. Learn how to automate your Power Platform administration by using the API directly—without any wrappers—and leveraging both delegated and application authentication while limiting scopes to reduce the attack surface!

Beyond the Agent: Intelligent Workflows, or How GitHub Copilot Harness Finally Handles Complex Processes

There’s a lot of talk about agents. Much less is said about what happens when the process takes three weeks, passes through five systems, involves four human approvals, and leaves no room for error. A conversational agent responds. An Intelligent Workflow, on the other hand, orchestrates, remembers, resumes after an interruption, and provides accountability. This session explores the least-explored—and richest—part of the new Harness: orchestration. How can we balance what must remain deterministic (rules, validations, compliance) with what benefits from a probabilistic approach (classification, extraction, decision-making under ambiguity)? Where should we draw the line—and why is drawing it in the wrong place the leading cause of failure for AI projects in production? On the agenda: orchestration patterns that handle the workload (long-running workflows, state and memory management, recovery points, escalation to humans, failover), the respective roles of Harness, Copilot Studio, and Power Automate in an enterprise architecture, and concrete criteria to help you determine whether your process falls under the category of an agent, an intelligent workflow, or both. You’ll leave with a “agent vs. intelligent workflow” framework, architectural patterns you can implement as early as Monday, and a clear understanding of how orchestration is transforming the governance of your business processes.

RBAC in Dataverse: When “Looks Secure” Is a Compliance Issue

Description: RBAC in Dataverse is often misunderstood. Not because it is complicated, but because it is usually explained in terms of roles and UI settings—rather than actual access behavior. Many solutions appear secure. Users see only “their” records. Buttons are hidden. Views are filtered. From a compliance perspective, this seems reassuring—until it isn’t. In this session, we’ll examine how access control in Dataverse behaves when solutions move beyond demos and into real, multi-user scenarios. We’ll explore why UI-based access logic creates a false sense of security, how it fails under collaboration requirements, and why this gap often becomes a compliance issue rather than a functional bug. Using practical, demo-driven examples, we’ll focus on Owner Teams and Access Teams as the mechanisms Dataverse provides to model access correctly: ✅ when records should be owned by teams rather than individuals ✅ how to manage dynamic, record-level access without manual sharing ✅ how access changes automatically as users join or leave a team You’ll learn: 💡 what RBAC in Dataverse actually enforces 💡 why UI logic does not meet compliance expectations 💡 how Owner Teams and Access Teams reduce over-permissioning 💡 how to design access models that remain defensible during audits No deep security theory. No “checkbox compliance.” Just practical access patterns that hold up when someone finally asks, “Who can see this data—and why?” (And yes, RBAC sometimes stands for Really Bad Access Control—especially during audits)

Studio Tweaks: Take Your Power Apps Studio to the Next Level!

During this session, we’ll introduce you to Studio Tweaks: an amazing plugin that will transform your Power Apps development studio by adding everything Microsoft hasn’t been able to provide yet. You’ll find: - a Formula Bar 2.0 (collapse/expand code, move to another screen, predefined and custom themes, and tons of super-handy settings) - a LIVE (detachable) explorer for collections and all your variables - an ultra-detailed and customizable documentation generator - an SVG integrator/editor - accessibility tools (contrast testing, visual impairment simulators, etc.) - a system for identifying and centralizing annotations in the code (TODO, FIXME, etc.): create your own keywords - a favorites bar with your favorite controls to insert into your screens - and lots of other little extras (universal color picker, advanced search, favorites bar to insert your favorite controls, etc.) You’ll never develop the same way again!

Power Platform for the Planet: Carbon, Measured, Corrected

44.3°C. Pissos, June 23. France’s hottest day since 1947, during France’s hottest summer since records began. Two thousand excess deaths in a single heat wave. Scientists say June would have been virtually impossible without us. So how did we get here? Fuel burned by vehicle fleets. Electricity powering plants. Freight moving goods. Greenhouse gas emissions, produced by companies—most of which still track those emissions in a spreadsheet, if at all. Luckily, Microsoft saves the day. This session shines a light on a product that deserves far more attention than it gets: Microsoft Sustainability Manager. Environmental data in one place, emissions calculated according to the GHG Protocol, an audit trail behind every number, and the disclosures that customers and regulators keep asking for. Then AI came along and changed the game. We’ll walk through everything Microsoft offers today: Copilot within Sustainability Manager, the Sustainability Insights agent, the CBAM estimator, and the brand-new agent feed running on the Power Apps MCP server—where agents stop reporting and start taking action, with a human keeping an eye on things. Come see a solution you’ve probably never even opened. Bring nothing but curiosity.

Generative pages: When? What? Why? How?

There’s a lot of talk about “vibe coding,” code apps, and managed apps, but what about Generative Pages—which are already part of the landscape and in general availability (GA)? Let’s take a look together at what they can do and when it makes sense to use them. And most importantly, why Generative Pages still have a role to play!

From Apps to Agents: Building MCP Apps in Copilot Studio

Copilot Studio is introducing MCP Apps, a new way to bring application-like experiences directly into agent workflows. This session will demonstrate how we're evolving Copilot Agent Kit applications into intelligent agent experiences powered by MCP Apps. We'll cover the architecture, design considerations, and the opportunities MCP Apps create for delivering richer, more interactive experiences within Copilot Studio

Governing Computer-Using Agents (CUA): A KYC Automation Case Study in Copilot Studio

Many organizations struggle to automate KYC processes because the information they need is scattered across identity documents, emails, web portals, and legacy applications that do not expose APIs. While traditional automation works well for structured systems, real-world compliance workflows often require navigating fragmented information, interpreting documents, and making risk-based decisions. This is where Computer-Using Agents (CUA) come in. In this session, we’ll explore how to design, implement, and govern CUA-powered automation in Microsoft Copilot Studio using a real-world KYC onboarding scenario. Rather than positioning CUA as simply the next generation of RPA, we’ll examine where it fits within an enterprise architecture, when it adds value beyond APIs, and what governance controls are required to operate safely in regulated environments. You’ll learn: • When to use CUA versus traditional integrations and automation approaches • How CUA can automate KYC workflows involving identity documents, external registries, legacy systems, and fragmented data sources • How to establish governance through confidence thresholds, human-in-the-loop escalation, audit trails, and decision traceability • How to implement a governed solution using Copilot Studio, Power Automate, Dataverse, Microsoft Entra, and Microsoft Purview Through a live demonstration of a KYC triage agent, we’ll walk through the complete journey from document intake and verification to risk assessment and escalation, showing how organizations can automate complex compliance workflows while maintaining transparency, accountability, and regulatory alignment. Key Takeaways 1. Understand the scenarios where CUA delivers value beyond APIs and traditional RPA. 2. Learn why KYC and compliance workloads are ideal candidates for CUA-driven automation. 3. Translate regulatory and governance requirements into practical Copilot Studio architecture patterns. 4. Implement an end-to-end pattern for document processing, verification, risk scoring, auditability, and human oversight. As the saying goes: “Automation without governance is just chaos at a faster pace.”

From Idea to Implementation: Demystifying New AI Tools for Makers

Generative AI is profoundly transforming the way we design and develop Power Platform solutions. Between Plan Designer, Code Apps, Vibe Coding, GitHub Copilot, and the various AI assistants built into Microsoft, it can sometimes be difficult to understand the role of each tool, how they complement one another, and their limitations. This session aims to demystify these concepts through hands-on demonstrations and real-world use cases. We’ll explore how to turn a simple business idea into a functional application by leveraging the capabilities of the various AI assistants available today. The goal isn’t to compare features on paper, but to answer a simple question: Which tool should you use for which need?

UDF / Power BI is scaling up.

UDFs aren’t just changing the way we write DAX. They’re also changing the way AI will be able to generate it in the future. For years, DAX developers have been copying and pasting hundreds of lines of code, creating countless nearly identical measures, and dreaming of a truly modular approach. Those days are over. In this session—which is 100% demonstrations—we’ll push User-Defined Functions to their limits: abstraction, reusability, advanced business logic, optimization, and next-generation Power BI model architecture. Using a common thread and real-world examples, we’ll explore, among other things: creating true DAX function libraries with namespaces; the dynamic generation of reusable calendar dimensions spanning multiple fiscal years; advanced windowing techniques using OFFSET, WINDOW, and INDEX to build generic time-series comparison calculations; the creation of DAX-driven graphical components using dynamic SVG (KPI charts, visual indicators, themes, and reusable components); the standardization of data exports using ExecuteQueries, Power Automate, and DAX pagination; function governance via TMDL, Git, and Power BI Projects. We’ll also explore how the VAL, EXPR, ANYVAL, and ANYREF parameters enable us to write functions capable of adapting to any business context. And most importantly, we’ll tackle a rarely discussed topic: why UDFs are likely the missing link between Power BI and generative AI. Because while AI can generate code, it excels when it can rely on standardized, documented, and reusable business abstractions. No marketing slides. No endless PowerPoint presentations. Just code, live demos, real-world business scenarios, and a vision of what DAX development will look like in the age of artificial intelligence. Get ready to see DAX code you’ve probably never written before. Before UDFs, we shared DAX patterns. With UDFs, we now share code. And soon, libraries of knowledge that AI can leverage.

Building the Bridge: Bidirectional AI Between Dynamics 365 and Copilot Using Xrm.Copilot

For years, AI in Dynamics 365 was confined to a side panel: helpful for answering questions, but disconnected from the app your users actually work in. That barrier has been broken down. The new Xrm.Copilot API interface in model-driven apps creates a true two-way bridge: your app can send context and requests to Microsoft 365 Copilot, and your agents can reach back into the app to open records, navigate views, render the UI, and suggest actions. The app and the agent finally share the same state. In this session, we’ll build that bridge live, following a single scenario from start to finish: a collections workspace for finance teams, delivered as a model-driven app using Business Central virtual tables, so every customer, invoice, and ledger entry on screen is live ERP data. DEMO: A collections specialist opens a customer record with overdue invoices. The app seamlessly shares what’s on screen with Copilot, so the conversation begins with a solid foundation in that specific customer, view, and filter. No copying and pasting, no need to explain context to a chatbot. With a single click, the agent can ask Copilot to analyze payment behavior. Copilot analyzes the open invoices and responds in the panel, but it doesn’t stop at text: it opens the critical invoices in side panes and provides the specialist with a prioritized, filtered list of exactly what to follow up on today. The agent is now using the app. Then comes the moment that changes how you think about UI: the specialist asks Copilot to build a dashboard showing overdue amounts by customer and aging bucket, and it appears right inside the app itself, rendered using the same components the agent uses in their own pane, with both interfaces kept in sync through shared state. ARCHITECTURE: We dive deep by explaining how a PCF control, a declarative agent, and an MCP Apps server work together to deliver a single experience across two interfaces. We conclude with a clear overview of what is generally available, what is still in preview, the licensing requirements, and the challenges you’ll encounter on the path to production. KEY TAKEAWAYS: - A comprehensive overview of Xrm.Copilot capabilities: what is generally available, what is in preview, and when to use each - The working bridge pattern: PCF, declarative agent, and MCP Apps server connected through prompts, context, actions, and shared state - How to build agents on top of ERP data using Business Central virtual tables in Dataverse - Generative UI in model-driven apps: dashboards and components your users create by requesting them - A production readiness checklist: licensing, governance, and the pitfalls that only become apparent when you actually build this

Battle: CodeApps vs. VibeCoding vs. AI-Assisted Development

Three approaches and three visions of today’s development go head-to-head on stage. Between CodeApps, VibeCoding, and AI-assisted development, which method best addresses your challenges? Hear the arguments, real-world feedback, and discussions with the audience: come help decide between these approaches and leave with a clear vision to make your own choices.

From Prototype to Production: Build, Test, and Manage Agents with the Copilot Agent Kit

Building an agent in Copilot Studio takes just a few minutes. The hard part is creating one that you can confidently test, deploy, and manage. The Copilot Agent Kit is a free, Microsoft-maintained toolkit that installs in your own tenant and bridges that gap. In 20 fast-paced minutes, we’ll explore the capabilities that developers rely on every day: automated batch testing with AI-graded rubrics, the Agent Debugger, Prompt Advisor, Conversation KPIs, and the Agent Insights Hub; the Compliance Hub for governance at scale; and the Agent and Component Libraries for reusable building blocks. You’ll leave knowing exactly which tool solves which problem—and how to get the entire kit up and running today.

Let's Take a Look at Copilot Managed Runtime—Is This the End of the Power Platform "as We Know It"?

What if your company’s next business app weren’t a Canvas app, a model-driven app, a CodeApp, or even an app built in Power Apps? With Copilot Managed Runtime, Microsoft is introducing a new hosting platform for apps—whether AI-generated or hand-coded—that run in the Microsoft 365 tenant under IT governance. No mandatory development studio: you design, code, and deploy. And yet, the foundation is the one we’ve known for over a decade: Dataverse, environments, connectors, DLP, Entra. Perhaps that’s the real masterstroke: changing everything on the front end without touching what gives CIOs peace of mind. This session is an introduction to the platform, complete with demos. For citizen developers, you build apps using Copilot Code and Copilot Studio, without writing a single line of code. For professionals, you switch to the SDK, the CLI, and Microsoft skills for managed apps: your favorite IDE and code editor. The developer workflow is changing radically: a Git repository managed by the platform or your own GitHub repository, a Git-native approach, and “just-in-time” builds. This is a true upheaval of ALM as we know it. We’re not going to write off Power Platform: the foundation we’ve been building for years is well on its way to enduring. But we’ll have to ask ourselves honestly whether Copilot Managed Runtime isn’t becoming the successor to the visible part—the part that users see.

Dataverse Business Skills & Semantic Model: Apply Your Business Rules to All Your AI Agents

With the proliferation of approaches to integrate Dataverse and D365 with generative AI (custom agents, MCP servers, Copilot Studio), a crucial architectural question arises: how can you ensure that all of these different agents behave consistently and strictly adhere to your business rules or project governance guidelines? Microsoft may have just addressed this issue. Thanks to the preview release of Business Skills and the Semantic Model, it’s now possible to enforce a unified set of guidelines across all your agents—whether they’re custom-developed, connected via MCP, or orchestrated within the Copilot Studio ecosystem. And you can do this without writing a single line of code. This is precisely the essential technological building block needed to reassure decision-makers and address skeptics who still had doubts about the governance and reliability of Gen AI in the enterprise. Since these features are still in preview, this session gets right to the point. Together, we’ll explore their target use cases and configuration, then move on to a live demo to see how to manage and secure your agents’ behavior directly from Dataverse.

Agent FinOps: Gaining Technical Control Over Copilot Studio Usage

A Copilot Studio agent can be built in a few hours, but its cost is determined at the design stage. This session details what actually consumes resources (generative responses, actions, knowledge sources), compares two architectures of the same agent along with their costs, and then shows how to manage resource consumption from a technical perspective: capacity allocation by environment, monitoring, and alerts. You’ll leave with a method for estimating the cost of an agent before deploying it.

Play It (W)right: Automate Your End-to-End Tests in the Power Platform

Modern Power Platform solutions are becoming increasingly complex, making reliable end-to-end testing more important than ever. This session helps Power Platform teams adopt Playwright as their primary testing framework. You'll learn about Playwright architecture, browser automation, accessibility testing, device emulation, screenshot and video capture, and integration with GitHub and Azure DevOps pipelines. We'll also explore how AI and Playwright work together. Through Playwright MCP workflows, Playwright Agents, and the Playwright CLI with skills, you’ll see how intelligent automation can accelerate test creation, maintenance, and diagnostics. Key takeaways - Build scalable Playwright test suites for Power Platform solutions, as well as modern web applications such as Dynamics 365 and Business Central. - Apply practical patterns for accessibility testing, device emulation, visual validation, and ALM integration. - Understand how AI-enabled Playwright workflows can accelerate test development, maintenance, and troubleshooting.

WIQD: How to Make WorkIQ and Plugins Accessible.

WIQD (pronounced "wicked") is the developer kit that unlocks the full extensibility of M365 Copilot. In this session, we’ll introduce the dev kit and show how it transforms our coding agents into an interface for connecting an agent to WorkIQ, creating an advanced plugin for Cowork and Copilot, or even taking it a step further by developing an MCP Apps server that can be used within your agents.

Claude Code Apps - Live!

Session on building Code Apps using an agent-based harness in Claude Code, based on insights from real-world production applications. Session outline: - What is an agent-based harness? - Example of a harness for Code Apps - Demo: From brief to mockup - Demo: From mockup to app

Secure Power Platform Architecture with Managed Identity and VNet

Security isn't optional when scaling Power Platform enterprise solutions—it's fundamental. This session explores how Managed Identity and Virtual Network (VNet) integration transform authentication, data access control, and network security for production-ready architectures. Learn how to securely connect Power Platform services to Azure resources without storing secrets, apply least-privilege access patterns, and eliminate the exposure of your Azure resources to the public internet using private endpoints. Through practical implementation patterns and live demos, we’ll cover best practices for securing Dataverse plugins and integrations with Azure services. Walk away with actionable blueprints to secure your Power Platform environment—no more secrets in environment variables, no more public endpoints.

All Brains, No Hands? Custom Connectors and Copilot Studio

Your Copilot Studio agent can reason through a problem and plan three steps ahead—provided it has enough tokens to use. Usually. However, it still can’t create the order or check inventory levels because it has no way to access the custom systems your business actually runs on. We’ll walk through two ways to give it a hand—pun intended: an MCP server or a custom connector. Then we’ll dive into custom connectors using real-world examples: how to build them (authentication, actions, triggers, policies, custom code), how to test them, and how to release them with confidence. To wrap things up, we’ll look at how they’re used: called directly by the agent, or embedded in a flow or app that locks down the logic for deterministic execution. A session for makers, developers, and architects building agents that connect!