Arize:Observe 2025 Keynote: Introducing Alyx & ADB

Explore the future of AI engineering, agent infrastructure, and evals in this Arize Observe 2025 keynote hosted by Arize AI founders Jason Lopatecki and Aparna Dhinakaran.

Chapters:
03:00 - Autonomy in Products: From Tab Completion to Task Solving
05:00 - The Challenge of Agent Context Management
06:00 - Evaluating, Controlling, and Improving AI Agents
06:46 - Introducing Alyx: Arize’s Contextual Agent for AI Engineers
08:01 - Alyx Demo: Trace Troubleshooting + Hotkey Workflows
10:20 - Visual Agent Debugging with Agent Flow Tracing
11:00 - Multi-Agent Architectures & Debugging Hand-offs
12:30 - Multi-Agent Tracing Product Launch
13:30 - Session & Trajectory Evaluations: Measuring Full-Agent Journeys
14:50 - Prompt Learning: Optimization Powered by Annotations & Feedback
16:20 - Prompt Playground Views: Shareable, Saveable IDE for Prompts
17:00 - Alyx MCP Assistant: IDE-Native Debugging Tools
18:00 - Commitment to Open Source: Phoenix Upgrades
19:00 - Multi-User Phoenix Cloud, Cost Tracking, and Bedrock Support
21:00 - One More Thing: Introducing ADB – Arize’s OLAP Engine for AI Data
22:30 - Inside ADB: Real-Time Ingest, Petabyte Scale, and Sub-Second Queries

This year’s keynote introduces Alyx, a redesigned in-product AI agent that brings rich context, native trace troubleshooting, and IDE integrations to your debugging workflows—alongside powerful new capabilities in multi-agent visualization, session-level evaluation, and prompt optimization.

???? From observability to real-time prompt tuning, this keynote is your front-row seat to the next decade of building with AI agents.

???? Featured Product Drops:

Alyx (Copilot v3): A context-rich agent that helps you diagnose issues on the fly—invoked with a hotkey (cmd+L) and integrated into Cursor or Claude Code via MCP.

Agent Flow Tracing: Visualize LLM decision paths and trace execution logic

Multi-Agent Tracing: Inspect agent handoffs and orchestration bottlenecks

Session & Trajectory Evaluations: Measure coherence, memory, and problem-solving steps across full conversations

Prompt Learning: Optimize prompts using human or LLM feedback—not just scalar scores

Prompt Playground Views: Save and share side-by-side prompt experiments

Alyx MCP Assistant: Access all of Alyx’s skills directly from Cursor, Claude Code, and more

Phoenix Upgrades: Multi-user collaboration, cost tracking, and Bedrock model support

ADB (Arize Database): A blazing-fast, scalable OLAP engine purpose-built for AI workloads

???? Whether you're building copilots, customer support agents, or research assistants, this keynote showcases the latest tools to help you evaluate, debug, and optimize LLM-powered systems at scale.

???? Read More:
Full product recap:
https://arize.com/blog/observe-2025-releases/

Inside ADB — Arize’s OLAP engine:
https://arize.com/blog/introducing-adb-arizes-proprietary-olap-database/

#AIEngineering #LLMops #AgentOps #Alyx #ADB #PromptEngineering #AIObservability #PhoenixAI #ArizeObserve #OpenSourceAI