CollinBrian explores how ambient AI agents—powered by change-driven architecture and tools like Drasi—can move beyond traditional chat interfaces to deliver real-time, reactive intelligence directly into workflows.

Beyond the Chat Window: How Change-Driven Architecture Enables Ambient AI Agents

Introduction

AI agents traditionally work via chat interfaces: you ask, they answer. But the next frontier is “ambient agents”—AI operating autonomously in the background, detecting real-world changes and responding instantly. This article explores this paradigm, the infrastructural challenges it brings, and how new tools are overcoming those hurdles.

Ambient Agents vs. Conversational AI

  • Conversational AI: Follows the familiar request-response cycle (user asks, agent answers).
  • Ambient Agents: Monitor streams of events, maintain context, and react without user prompting. They excel at:
    • Real-time infrastructure monitoring
    • Automated remediation
    • Continuous context maintenance for applications

The Problem: Real-Time Change Detection

Ambient agents must ingest and act upon a constant, real-time flow of data from multiple sources. Traditional approaches struggle with:

  • Unscalable polling (wastes resources, misses events)
  • Rewriting legacy systems to emit events
  • Managing unreliable and inconsistent webhooks

The Solution: Drasi

Drasi is a change detection engine that acts as the “sensory system” for your AI agents. Key features:

  • Sources: Connects to databases (PostgreSQL, MySQL, Cosmos DB), Kubernetes, EventHub, and more.
  • Continuous Queries: Uses graph-based queries (Cypher/GQL) to monitor for relevant changes.
  • Reactions: Triggers actions or notifications when conditions are met.
  • Goes beyond “something changed”—understands what changed and why it matters, including detection of lack of change.

Integration: langchain-drasi

The langchain-drasi integration bridges Drasi’s event detection with LangChain’s agent frameworks. Agents can:

  • Discover available change queries
  • Read and act on results in real time
  • Subscribe to push updates, integrating events into agent memory and workflow

Example Code

from langchain_drasi import create_drasi_tool, MCPConnectionConfig

mcp_config = MCPConnectionConfig(server_url="http://localhost:8083")
drasi_tool = create_drasi_tool(
    mcp_config=mcp_config,
    notification_handlers=[buffer_handler, console_handler]
)

Notification handlers make it easy to direct reactions into buffers, agent memory checkpoints, or logs.

Concrete Example: AI NPC Seeker Agent

A multiplayer game logs player positions to PostgreSQL. An AI NPC agent uses Drasi queries to:

  • Detect stationary players (no movement for >3 seconds)
  • Spot “frantic” players (multiple moves in under a second)

Drasi’s continuous queries handle event detection efficiently. The LangChain agent subscribes to these, evaluates targets, plans moves, and acts—all without polling.

The Big Picture: Change-Driven Architecture

This approach establishes a new pattern: AI solutions that respond to real-world changes, not just user prompts. Practical potential includes:

  • Smart city management
  • Disaster response
  • Adaptive supply chain logistics
  • Real-time infrastructure protection

Next Steps

This post appeared first on “Microsoft Tech Community”. Read the entire article here