A language model does not have its own memory. It can only respond based on the the messages and context provided to it in each conversation. Providing the right context is critical to getting results that feel personalized and intelligent. This is where memory comes in. But what does memory look like in an AI agent? How can we build systems that remember user preferences across conversations, without overwhelming users with questions or forcing them to repeat themselves? In this post, we will explore short-term and long-term agent memory.

Tools are the hands of the AI agent, enabling it to perform actions beyond generating text. There are tools for querying databases, calling APIs, sending emails, and a host of other capabilities. Tools may sometimes perform actions that change system state or need access to sensitive data. This is where it is critical to stay in control to ensure security and compliance. In this post, we'll explore how to build secure approval workflows using Microsoft Agent Framework and AG-UI, ensuring that critical actions require explicit user consent before execution.

While Language Models excel at generating human-like text, their responses can be unpredictable in format and structure. The format is important when the responses need to be consumed by other downstream systems. JSON has always been a popular choice for structured data interchange. In this post, we will bridge the gap between conversational AI and reliable system integration by using JSON Schema to enforce structured outputs in AI agents.

Modern AI Agents are becoming increasingly powerful, but their true potential is unlocked when they can interact with external systems. In my previous post, I discussed how to create MCP (Model Context Protocol) servers and the advantages of using MCP to connect AI models to external tools and data sources vs. traditional function callings. In this follow-up post, I will dive into using MCP tools with Azure AI Foundry Agents.

AI coding assistants have become an essential part of my daily workflow. Among the various many available, GitHub Copilot stands out as the most powerful and versatile. GitHub Copilot integrates with Visual Studio Code and other popular IDEs, providing context-aware suggestions that help me write code faster and smarter. In this post, I will share my personal workflow with GitHub Copilot — why it has become such a big productivity boost, and how it goes far beyond simple code completion to support many other tasks.