From Figma design to the PowerApps with Lukas Pavelka [MVP]

Development is changing faster than most teams can process. A few years ago, building enterprise applications meant long development cycles, hand-coded UI layers, endless testing loops, and massive backlogs between design teams and developers. Now AI agents can write code, generate layouts, repair syntax, optimize workflows, and even help translate entire applications into more than one hundred languages. But that shift creates a new question: If AI can generate applications faster than ever before, what actually separates good development from dangerous development? In this episode of the M365 FM Podcast, Mirko Peters sits down with Microsoft MVP Lukas Pavelka to explore the intersection of Figma, PowerApps, AI-assisted coding, Power BI, and the rapidly changing future of enterprise application development. The conversation goes far beyond low-code hype. This episode explores what really happens when AI agents enter the development lifecycle, how Figma is evolving into a complete ecosystem, why governance and security still matter deeply in AI-driven coding, and how developers can use tools like Copilot, Claude, GitHub Copilot, and vibe coding without losing control of their own codebase.FROM JAVA DEVELOPER TO FIGMA AND POWERAPPS CREATOR Lukas Pavelka started as a traditional Java developer more than twenty years ago before eventually transitioning into Power Platform development, automation, and AI-assisted application design. The turning point came through design. After discovering Figma through his wife’s design work, Lukas realized there was a major gap between beautiful design systems and practical PowerApps development workflows. That led to the creation of his PowerApps for Figma plugin, designed to help Power Platform developers move much faster between design and implementation. Today, Lukas develops multiple products focused on bridging design, automation, AI, and low-code development, including:PowerApps for FigmaPower BI for FigmaMy Bot Admin for Telegram automationThe discussion explores how these products evolved from internal productivity ideas into community-focused tools aimed at helping developers, makers, and Power Platform teams reduce repetitive work and improve enterprise UI quality.WHY FIGMA IS BECOMING MUCH BIGGER THAN DESIGN One of the most fascinating parts of this episode is the discussion around Figma’s evolution. Lukas explains why Figma is no longer just a design platform. It is becoming a complete ecosystem that increasingly overlaps with development, prototyping, presentations, AI-assisted workflows, and enterprise application delivery. The conversation covers:Figma design systemsReusable component librariesPowerApps UI translationYAML exportComponent variantsMulti-language enterprise appsDesign consistency across projectsLukas also explains how his plugins allow Power Platform developers to create scalable design systems that can be reused across enterprise projects while dramatically reducing repetitive UI work. The discussion highlights a major shift happening inside enterprise development: Good UX is no longer optional. Organizations increasingly realize that internal business applications must feel modern, intuitive, and scalable if they want employees to actually use them effectively.AI, VIBE CODING, AND THE REALITY OF MODERN DEVELOPMENTThis episode dives deeply into AI-assisted development and the rise of “vibe coding.” Lukas shares practical experiences using GitHub Copilot, Claude, Visual Studio integrations, AI agents, and prompt-based coding workflows to accelerate development. But the conversation stays grounded in reality. One of the strongest themes throughout the episode is that AI coding still requires strong technical understanding. Lukas explains why developers cannot simply rely on AI-generated code without understanding architecture, debugging, security, versioning, and governance. The discussion explores:Prompt engineering for developersAI-assisted debuggingModel selection strategiesToken cost managementVersioning challengesSecure coding practicesMCP and Model Context ProtocolAI coding limitationsA major insight from the episode is that AI coding works best when prompts stay highly focused and scoped to one specific task at a time. Broader prompts often cause AI agents to rewrite working code unnecessarily or introduce instability into existing projects. The episode also explores how AI development changes the role of the developer itself. Instead of writing every line manually, developers increasingly supervise, guide, validate, secure, and orchestrate AI-generated output.THE BUSINESS REALITY OF AI DEVELOPMENT The conversation also moves into the economics behind AI-assisted development. Lukas and Mirko discuss token costs, cloud compute limitations, GPU demand, electricity consumption, and the growing operational cost of running large-scale AI systems. The episode examines:Claude pricingGitHub Copilot limitsAI token consumptionGPU infrastructureElectricity challengesAI model specializationCloud economicsOne particularly interesting part of the discussion focuses on how different AI models perform better for different development tasks. Some models perform better for frontend design work, others for deeper reasoning, debugging, or enterprise coding scenarios. This creates a new challenge for developers: Understanding not only how to code, but also which AI model to use for which type of work.SECURITY, GOVERNANCE, AND THE RISKS OF AI CODING As AI-generated development accelerates, governance becomes increasingly important. Lukas explains why developers still need to understand exactly what their code is doing, even when AI agents generate large portions of it automatically. The episode explores the growing risks around:Security vulnerabilitiesPoor governanceExposed repositoriesUnsafe promptsWeak versioning practicesAI-generated technical debtOne of the strongest warnings throughout the episode is simple: AI can accelerate bad development just as easily as good development. Without proper architecture, security awareness, governance structures, and development knowledge, organizations risk creating large amounts of insecure code much faster than before.WHAT COMES NEXT FOR AI DEVELOPMENT The future discussed in this episode moves beyond simple text prompts. Lukas explains why voice-driven development, AI skills, reusable agent capabilities, and contextual AI orchestration are becoming the next major wave in application delivery. The discussion explores how future AI systems may:Understand spoken instructionsBuild applications conversationallyReuse trained development skillsOrchestrate workflows automaticallyConnect through MCP serversGenerate full enterprise UI systemsAt the same time, both Lukas and Mirko emphasize that strong development fundamentals remain essential. The tools are changing rapidly. But architecture, security, UX thinking, governance, and operational understanding still matter moBecome a supporter of this podcast: https://www.spreaker.com/podcast/m365-fm-modern-work-security-and-productivity-with-microsoft-365--6704921/support.