Microsoft and Mastercard Are Racing to Build AI That Thinks Like a Business Partner

Microsoft and Mastercard are at the forefront of developing agentic AI systems, demonstrating the growing significance of autonomous artificial intelligence in business and commerce. Agentic AI refers to systems that can independently set and pursue goals, make decisions based on real-time data, and continuously learn and improve from each interaction. These systems are not static tools but dynamic agents capable of adapting to changing conditions and coordinating with other systems or humans to achieve complex tasks.

One of the most notable agentic AI developments is Microsoft’s AutoGen framework, which simplifies the orchestration of multiple AI agents. AutoGen reduces complex coding tasks to a fraction of the effort typically required by abstracting away the technical details of message routing, API calls, and state management. This tool is particularly useful for enterprises aiming to integrate AI into workflows that require collaboration between multiple specialized agents. For instance, AutoGen can be used in utility analysis systems to automate the identification of infrastructure issues, compliance checks, and strategic planning. By enabling each agent to perform its specific role without interference, AutoGen ensures a clear separation of concerns, making it easier to manage complex multi-agent systems [2].

Mastercard has also embraced agentic AI with the launch of its Agent Pay technology, a pioneering system designed to enable AI shopping agents to make purchases on behalf of consumers and businesses. This innovation represents a significant shift in the retail sector, allowing users to delegate shopping tasks to AI agents that can search for products across multiple platforms, compare options, and complete transactions automatically. This level of automation enhances efficiency and personalization, as AI agents can remember user preferences and past purchases to make more informed decisions. However, this new form of commerce also introduces challenges, such as the need for clear user permissions, transparent decision-making processes, and robust governance frameworks to address potential issues like chargebacks or consumer responsibility in cases where purchases go wrong [3].

Despite the growing interest in agentic AI, its implementation is fraught with challenges. According to research, over 90% of agentic AI projects fail to deliver expected returns within the first six months. One of the primary reasons for this high failure rate is the tendency to treat agentic AI like traditional automation systems. Unlike robotic process automation (RPA), agentic AI systems require ongoing training, refinement, and continuous oversight. Another common pitfall is the lack of clear success metrics. Many projects launch with vague goals like “improve productivity” or “reduce costs,” without defining specific, measurable outcomes. This makes it difficult to evaluate the effectiveness of the AI agent and justify the investment [4].

Successful agentic AI implementations share several key characteristics. First, they start with clear process documentation, ensuring that teams have a deep understanding of the workflows they aim to automate. Second, they design for oversight rather than full autonomy, creating structured workflows with clear escalation paths and human checkpoints. This approach helps maintain quality control while still leveraging the efficiency of automation. Third, successful projects measure everything—from agent performance to business outcomes—enabling continuous optimization. Finally, they plan for iteration from the outset, allocating resources for post-launch improvements and partner with platforms designed for production environments [4].

One of the most notable success stories is Avi Medical, a healthcare provider that reduced costs by 93% through the deployment of multilingual AI agents. These agents were integrated into existing systems and designed to handle 81% of common patient inquiries while escalating complex cases to human agents. The result was a 87% reduction in median response times and a 9% increase in patient satisfaction. This case highlights the potential of agentic AI to transform industries by combining automation with human oversight, ensuring both efficiency and quality [4].

As agentic AI continues to evolve, its applications are expected to expand beyond retail and healthcare into areas such as customer service, software development, and content creation. The future of agentic commerce may see AI agents managing entire purchasing journeys, from product discovery to payment and delivery. However, for these systems to gain widespread adoption, it will be crucial to address concerns around transparency, accountability, and security. Establishing clear standards and frameworks will be essential for building trust among consumers, businesses, and regulators.

Source:

[1] The Best Trillion-Dollar Stock to Buy Now, According to... (https://www.fool.com/investing/2025/09/03/best-trillion-dollar-stock-buy-wall-street-nvidia/)

[2] Simplify Multi-Agent AI Orchestration with Microsoft AutoGen (https://www.annielytics.com/blog/ai/simplify-multi-agent-ai-orchestration-with-microsoft-autogen/)

[3] What is agentic commerce? Your guide to AI-assisted retail (https://www.mastercard.com/us/en/news-and-trends/stories/2025/agentic-commerce-explainer.html)

[4] Agentic AI in 2025: Why 90% of Implementations Fail (And...) (https://beam.ai/agentic-insights/agentic-ai-in-2025-why-90-of-implementations-fail-(and-how-to-be-the-10-))