2025 has become the year of the agent. I greatly appreciate the opportunity to attend the Microsoft Build conference to explore the vision and roadmap for AI computing that is slowly transforming enterprises. The Work Trend Index Report from Microsoft observes that 80% of business leaders expect AI to impact their business culture and processes in the next two years.
The early moments of the conference reflected on the big themes cutting across our industry and multiple companies. (AWS, Google, Microsoft, OpenAI) In the early days of the world wide web, Tim Berners-Lee and the semantic web founders explored the vision of organizing a world-scale knowledge graph using markup languages that specializes in representing ontologies, objects and their relationships. The early semantic web explored the potential for intelligent agents to create new knowledge through inference of structured ontology data. In 2025, with the advent of large language models, we will see this vision come to life while avoiding the complexity of formal ontology construction. As humans, we represent and pass on knowledge through language. LLMs explore the written word and become a tool for creating insight.
Moving beyond the base functions of language models, the industry has shifted attention to the creation of agents, computer programs that can accept tasks, plan work, adapt to changing context, execute tasks, and achieve practical goals. The AI genie has been released from it’s bottle. It feels like every business and technology leader seeks to understand and explore AI agent technology. Companies feel that AI gives them competitive advantage and helps teams minimize the tedious and boring tasks of their backlog.
One of my favorite sessions of the conference focused on the Microsoft Research perspective of antigenic computing. In the past few years, their team has reflected on the consequences of having a “society of agents.” How would your work change if you could hire a team of agents to help you accomplish your mission? Really enjoyed seeing vision from this session.
1. Human Guidance and Prompting
Agents rely on human input to shape their behavior, objectives, and outputs. Prompts help direct their responses, whether through natural language queries or structured commands. Human oversight ensures ethical and practical alignment, refining agent actions and adapting their learning.
2. Planning
This involves setting goals, prioritizing tasks, and sequencing actions to achieve desired outcomes. Agents use techniques like search algorithms, heuristics, and optimization strategies to map out decisions efficiently—whether that’s finding the best route in navigation or determining the next step in problem-solving.
3. Reasoning
Agents employ logical deduction, probabilistic inference, and contextual understanding to make informed decisions. Some use rule-based systems, while advanced models leverage machine learning and neural networks to predict and adapt. The ability to evaluate various possibilities and consequences improves their effectiveness.
4. Memory
Memory allows agents to store, retrieve, and utilize previous interactions and learned experiences. Some AI models have short-term memory, retaining information within a single session, while others with long-term memory recall details across multiple conversations or tasks. Memory enhances personalization and continuity in interactions.
5. Communication: Agent Communication & Emerging Open Standards
Agents exchange information through structured protocols, APIs, and natural language processing. Multi-agent systems collaborate to solve complex problems. Open standards, like AI interoperability frameworks and standardized APIs, ensure seamless integration between different systems, enabling broad connectivity and shared knowledge. In the Build Conference, Microsoft celebrated the exploration of agent to agent communication using Anthropics Model Context protocol and Google’s “agent to agent” protocol.
6. Tools
Agents leverage tools—whether built-in functionalities or external APIs—to enhance their capabilities. These can range from search engines, data analysis tools, image generation models, reasoning engines, and automation scripts. The ability to invoke tools makes agents highly adaptable and efficient.
In the Build Conference, Microsoft explored various aspects of Azure AI Foundry, their new platform as a service for making AI agents. Azure AI Foundry is a powerful agent-building tool within Microsoft’s Azure AI ecosystem, designed to help developers create, deploy, and manage AI agents efficiently. Unlike tools like open source agent frameworks, which focus on chaining language models with external data sources, Azure AI Foundry offers deep integration with Azure’s enterprise-grade AI services, robust security, and seamless scalability. This enables businesses to build sophisticated, reliable AI agents with enterprise-level governance, monitoring, and optimization capabilities—all within a unified Azure environment. There’s tons to unpack here. For now, you should check out this fun talk exploring how Scott Hanselman is using Foundry to streamline his podcasting management, planning, and show notes.
Make sure to check out this getting started guide:
Quickstart: Get started with Azure AI Foundry
This year really feels like a science fiction becoming real. Are we ready? How will our work forces adapt as AI displaces or replaces certain elements of work? While there continues to be great promise, we need to continue to figure out how to value people first in the new agent enabled world.
Leave a Reply