1. Arize AX
  2. Phoenix OSS
  3. Docs
  4. Learn
  5. Company
  6. Pricing
Sign up Book a demo

Resources

Featured report

How to evaluate whether your AI agents are actually improving

Measure AI agent improvement with task completion, cost per completed task, regression checks, and consistent before-and-after evaluations.

Sara Verdi 9 min read September 2026
Read the guide
Papers

AI governance in banking: How to scale AI agents

Read more
Guide

Arize AX or Arize Phoenix: When to use each or both

Read the guide
Papers

The agent reliability gap

Read more
See all resources
Latest resources

Practical guides, field notes & frameworks for reliable AI agents.

Code-first for engineers. Quality frameworks for product managers. Operating models for leaders.

See all resources
Guide

Build vs. buy AI observability tooling: a total cost of ownership guide

Should you build, self-host, or buy an AI observability platform? Compare engineering effort, infrastructure, maintenance, security, and total…

Read the guide
Guide

6 LLM gateways to consider for production AI in 2026

Compare LiteLLM, Portkey, OpenRouter, TrueFoundry, Kong AI Gateway, and Cloudflare AI Gateway on routing, caching, deployment, pricing, and…

Read the guide
Guide

Self-improving agents: what changes, what persists, and how to prove it

Learn how self-improving AI agents turn production evidence into persistent changes, and how to evaluate, validate, govern, and…

Read the guide
Guide

Agent reliability: how to measure and improve AI agents in production

Agent reliability is whether an AI agent consistently completes its task under real conditions. Learn the metrics, failure…

Read the guide
The Evaluator newsletter

The agent feedback loop, in your inbox.

New playbooks, field notes, and frameworks for building reliable AI agents.

Guides

Go deep, chapter by chapter.

Long-form handbooks you can read end to end, or drop into at the chapter you need.

Browse all resources
Handbook

What are AI agents? Architecture, tools & how they work

Learn what AI agents are, how they work, and how to build them. Explore agent architecture, tools, memory, orchestration, security, observability, and evaluation.

  1. 01 AI agent frameworks compared: LangGraph, CrewAI, AutoGen, and more
  2. 02 Agent observability: how to trace, debug, and improve AI agents
  3. 03 Agent evaluation: How to evaluate AI agents in production
  4. 04 Agent evaluation metrics: how to measure whether an agent works
Start the guide 4 chapters
Guide

The definitive guide to LLM evaluations

LLM evaluation: Get from pre-production to deployment with our definitive guide to LLM evaluation. Includes LLM eval types, use cases, templates and tips for continuous improvement.

  1. 01 LLM evaluation metrics: correctness, groundedness, RAG & agent scores
  2. 02 Pre-production LLM evaluation: datasets, synthetic data & benchmarks
  3. 03 CI/CD for LLM apps: experiments, regression tests & release gates
  4. 04 Production LLM evaluation: guardrails, online evals & monitoring
Start the guide 4 of 5 chapters
From the blog

The latest field notes and frameworks.

Explore the blog
Blog

TypeSafe’s Jev: Can decision models replace LLM judges?

TypeSafe’s Jev classifies, scores, and routes without generating text — up to hundreds of times cheaper than an…

Read the post
Blog

How to find and debug agent failures your evals are missing

Evals measure failures you know how to name. Arize Signal continuously reviews production traces to find recurring trajectory…

Read the post
Post

The future of AI operations teams

Your agents produce more traces and eval results than your team can read. SallyAnn DeLucia on how engineering…

Read the post
Customer lessons

Real teams, shipping AI.

See all case studies
Case Studies

How Booking.com scales AI observability with Arize

How Booking.com built a unified AI observability stack with Arize for agentic GenAI workflows and traditional ML —…

Read the story
Case Studies

How LG Uplus is building better AI customer service agents with evaluation-driven development

How LG Uplus uses Arize AX to build evaluation-driven AI contact center agents — combining production traces, user…

Read the story
Case Studies

How Tripadvisor is building the AI product development lifecycle for agentic travel

Tripadvisor VP of Data and AI Rahul Todkar on building a production AI lifecycle for traditional ML and…

Read the story
Compare

How Arize stacks up.

Side-by-side breakdowns for teams evaluating tracing, evals, and agent observability tools.

Arize vs. Braintrust

Compare Arize and Braintrust for AI observability, agent evaluation, tracing, production monitoring, deployment, and enterprise scale.

Read the comparison

Arize vs. Langfuse

Compare Arize Phoenix vs. Langfuse on open source tracing, agent evals, self-hosting, and pricing, plus when teams add Arize AX for production scale.

Read the comparison

Arize vs. LangSmith

Compare Arize vs. LangSmith for AI observability, agent evaluation, OpenTelemetry, failure discovery, self-hosting, and pricing to choose the right platform.

Read the comparison

Arize Signal vs. LangSmith Engine vs. Braintrust Topics

How each analyzes production traces, diagnoses failures, and closes the agent improvement loop.

Read the guide

Breaking down AI observability pricing

Compare AI observability pricing across LangSmith, Langfuse, Braintrust, Datadog, and Arize.

Read the guide

Arize alternatives: How the top AI observability and evaluation tools compare

Compare LangSmith, Langfuse, Braintrust, Helicone, Fiddler, and Phoenix on agent evals, tracing, self-hosting, and more.

Read the guide
Videos & talks

Demos, workshops & conference talks.

Watch on YouTube

An agent got the right answer the wrong way | Michael Grinich, WorkOS

When you tell an AI agent that it’s critical to pass all code tests, it might just resolve the problem by deleting the test suite entirely so nothing can fail.

Rise of the AI Engineer 2:36
Answer

Why do capable AI agents still fail in production?

Short answer

A capable model can still fail when the system around it breaks. Agents depend on context, tools, state, infrastructure, evaluation, and runtime controls. A strong model may even work around defects and return a plausible answer, making the system look healthier than it is.

Why it matters

The final answer only tells you part of the story. A successful output can hide bad retrieval, unnecessary retries, broken tool calls, lost state, rising cost, or unsafe actions. To understand reliability, teams need enough trace evidence to reconstruct the full trajectory and see how the agent reached its result.

The agent reliability feedback loop

  1. 1 Instrument complete traces
  2. 2 Inspect representative runs
  3. 3 Classify recurring failures
  4. 4 Turn failures into evals
  5. 5 Bring domain experts into review
  6. 6 Replay fixes, release, and monitor
Report

The agent reliability gap

Get the report

Don’t ship vibes.

Arize gives AI teams observability and evals to understand and improve agent performance.

Get started Book a demo