Balázs Horváth is the founder and lead architect of VisualLabs, a Microsoft business-applications consultancy focused on enterprise software, data, and AI. As AI makes code cheaper to produce, he argues, the harder engineering problem is deciding what deserves to be built.
Before founding VisualLabs, Horváth worked on enterprise resource planning and customer relationship management programs in the United States and the United Kingdom, including Microsoft transformation projects at KPMG and HCL. He began by testing and writing functional specifications, developing an approach to software delivery grounded in translating business requirements into instructions developers, consultants, and AI systems can execute.
His work spans Microsoft Dynamics 365, Power Platform, data architecture, and enterprise AI readiness. In his AI Engineer World’s Fair talk, he described an internal hackathon where 17 of 21 proposed AI agents were abandoned because they lacked business value or access to necessary data.
How he approaches enterprise AI
- Value → Architecture → Design. Horváth starts with the desired business outcome and existing process, examines supporting systems and data, and then designs the solution. He asks whose problem the application solves, how that person defines success, what might prevent adoption, and whether it improves a real decision.
- Story mapping for AI-assisted delivery. He breaks workflows into user stories specifying the intended user, need, purpose, and acceptance criteria, then prioritizes a minimum viable release. Keeping those requirements in repository-based Markdown gives coding agents concrete business context and testable specifications.
- Production adoption over polished demos. Horváth judges software by repeated meaningful use, not feature counts or impressive demonstrations, and advocates involving experienced technical specialists earlier in customer-facing decisions.
His analysis of Dynamics 365 and Power Platform applies the same thinking to platform selection, weighing business processes, integration, licensing, and implementation costs. His writing on modern data foundations for scalable AI emphasizes that enterprise AI depends on accessible, well-organized operational data.
