Microsoft recently unveiled Microsoft Discovery at its Build 2025 conference, a platform designed to harness the power of agentic AI to accelerate scientific discovery. According to the company, Microsoft Discovery aims to transform the entire research and discovery process from hypothesis formulation to candidate generation and simulation analysis. By enabling scientists and researchers to collaborate with specialized AI agents, Microsoft hopes to drive scientific outcomes with greater speed, scale, and accuracy, leveraging the latest advancements in AI and supercomputing. This announcement aligns with a broader trend in the tech industry, where major players are experimenting with AI's role in scientific research. For instance, Google has introduced its "AI co-scientist" to assist scientists in hypothesis creation and research planning. Similarly, companies like Anthropic, OpenAI, FutureHouse, and Lila Sciences have expressed optimism about AI's potential to revolutionize fields like medicine. However, critics argue that current AI systems are unreliable and have yet to prove their effectiveness in practical applications. Google's AI, GNoME, claimed to synthesize 40 new materials, but an independent analysis revealed none were genuinely novel. Drug discovery firms like Exscientia and BenevolentAI have also encountered high-profile clinical trial failures, casting doubts on the immediate benefits of AI in this domain. Despite these challenges, Microsoft is optimistic about the potential of Microsoft Discovery. The company aims to avoid the pitfalls of previous AI efforts by focusing on creating a versatile and extensible platform. Microsoft Discovery leverages the Model Context Protocol (MCP), a standard initially developed by Anthropic. MCP allows AI agents to interact with databases and services in a controlled manner, bypassing the need to navigate websites manually and reducing the potential for errors and unreliability. This protocol is crucial for building an "agentic web," where users can delegate tasks to agents, and these agents can perform actions on behalf of users, such as booking a hotel room or adding events to a calendar. The concept of the agentic web is rooted in the idea that agents can operate asynchronously, handling tasks while users focus on other activities. This asynchronous model contrasts with the current synchronous web, where users must actively engage with each task, such as completing a transaction on a Shopify storefront. MCP and the Natural Language Web (NLWeb) protocols enable websites to become more "agent-aware" by providing standardized ways to expose content and services to agents. For content providers, this could mean new avenues for traffic and engagement, allowing them to curate relationships with users more effectively. To illustrate the practical implementation of MCP and NLWeb, Microsoft demonstrated a low-cost, easy-to-integrate natural language search tool for websites. This tool, showcased in a demo with TripAdvisor, allowed the travel company to quickly implement advanced search capabilities, significantly improving user experience. The ease of integration and the ability to control how agents interact with the website are key features of this approach, as they address the business model concerns of content providers. Instead of simply hoping for organic search referrals, websites can now offer their content and services in a way that aligns with their own goals and preferences. However, the transition to an agentic web faces significant hurdles, particularly the need to reconcile with the existing dominance of platforms like Google and Facebook. Historically, these platforms have had to adapt to allow themselves to be searched and indexed by search engines, leading to increased exposure and traffic. Whether they will willingly open up to agentic AI is less clear, as they might perceive it as a threat to their own agent-based initiatives. Microsoft’s Chief Technology Officer, Kevin Scott, acknowledges this challenge but remains optimistic, believing that the technological and business incentives for adoption are strong. Scott emphasizes the importance of open protocols in enabling the agentic web, drawing parallels to the early days of the internet and the emergence of the web as the primary application platform on desktops. He argues that the current generation of AI technology is ripe for creating more sophisticated and useful agent systems. However, he also notes that there is a significant "capability overhang" with AI models—meaning they are more capable than currently utilized. Addressing this gap requires opening up the action space for agents and fostering an environment where entrepreneurs and developers can innovate freely. Scott's vision also touches on the ethical and economic implications of AI, particularly in the creative arts and media. While he is supportive of efforts to ensure fair compensation for creators, he believes that the broader benefits of AI, such as improving medical diagnostics for underserved populations, should not be overlooked. The key, he suggests, is to balance the needs of creators and consumers while ensuring that the technology remains accessible and beneficial. Industry insiders view Microsoft's initiative positively, noting that the company's commitment to open protocols and collaboration with other leading AI firms, including OpenAI and Anthropic, could drive widespread adoption. Despite tensions, Microsoft continues to work closely with OpenAI, contributing to the optimization of AI models and ensuring responsible deployment through joint safety boards. This partnership, albeit evolving, underscores Microsoft’s strategic focus on building a robust AI ecosystem. Microsoft’s efforts to redefine search and the web through agent-centric technology demonstrate a forward-thinking approach that could fundamentally change how information is accessed and used online. While the path to realization is fraught with challenges, the potential benefits in terms of efficiency, scalability, and innovation make it a compelling vision for the future of the internet and scientific discovery.