CrewAI、エージェント構築ツール「Crew Studio」をリリースしArize AXとネイティブ連携

CrewAI has launched Crew Studio, the Automated Agent Builder.

Through a native Arize AX integration, teams can send traces from Studio to Arize from the first run without adding custom instrumentation code. Builders and platform teams can then inspect agent behavior, evaluate quality, identify failures, and test improvements before redeploying.

What is Crew Studio

Crew Studio is a new building layer for agentic systems. It is designed to bridge a familiar gap in enterprise agent development: visual tools can become limiting as workflows grow more complex, while code-first frameworks may be inaccessible to the domain experts closest to the underlying business process.

Studio brings those workflows together across four stages:

Discover. A discovery module helps teams identify which agent use cases are worth building, grounded in patterns from real deployments rather than guesswork.

Build. Teams describe the system they want, and Studio turns that into a working architecture: agents, tasks, tools, models, flows, memory, and MCP connections. The output is informed by more than 700,000 patterns and architectures from billions of executions on the CrewAI platform.

Own. You own what gets built. Download the code, modify it, extend it, and move from visual building to code without starting over.

Run. Deploy to the same CrewAI AMP infrastructure your engineers already use, with governance, security, cost controls, and deployment paths attached.

Business teams can build while platform teams can stay in control.

Why observability has to be there from the start

Studio will put agent building in the hands of many more people. That’s the point. But it also means more agents in production, built by people who may never open the underlying code. When one of those agents misbehaves, someone has to answer what happened, which step failed, what it cost, and whether the fix worked.

That’s the gap this integration closes. Studio and AMP give teams traces, cost visibility, and governance in the same environment where agents are built and deployed. Arize AX adds a dedicated evaluation layer on top: online evals against live traffic, labeling queues, ground truth datasets, and experiments to validate a fix before it ships. Traces from every Studio deployment can land in Arize, where teams get span-level visibility into each agent run, latency and token cost breakdowns, and agent graphs that show how the system actually behaves in production.

Visibility is only the first step. Arize runs online evaluations against live traffic, so quality issues surface as they happen instead of when a user complains. Flagged traces route to labeling queues where human reviewers turn them into ground truth datasets. Those datasets feed experiments to test fixes before they ship.

The resulting feedback loop is straightforward: trace, evaluate, label, improve, redeploy. Every production run adds evidence that makes the next version better.

Built on OpenTelemetry standards

This integration works because of a capability Arize announced recently: native support for the OpenTelemetry GenAI semantic conventions, alongside OpenInference. Arize AX can ingest those traces directly and also supports applications instrumented with OpenInference semantic conventions.

To connect the platforms, configure Studio with Arize’s OTLP endpoint and authentication headers. No custom tracing code is required. Because the integration uses open telemetry standards, teams retain flexibility across frameworks, models, and observability backends.

How to connect Crew Studio to Arize AX

Connecting Crew Studio to Arize AX requires only a few configuration steps.

1. In Arize AX, go to Settings → API Keys. Copy your Space ID and create or copy an API key.

1. In CrewAI, go to Settings → Organization → OpenTelemetry Collectors, click Add Collector, and select OpenTelemetry Traces.

1. Configure the collector to emit traces to Arize:

Endpoint: https://otlp.arize.com/v1/traces

Service Name: your CrewAI service or project name

Custom headers:

arize-space-id: <your Space ID>

arize-api-key: <your API key>

1. Test the connection and save.

Once the collector is saved, traces from new agent runs appear in Arize under the configured service name. Teams can inspect span details, latency, token usage, and available cost data for each trace.

What customers are saying

PepsiCo, a mutual customer running agents in production with both platforms, put it this way:

“CrewAI and Arize are making it easier to build enterprise agents faster while providing the observability and evaluation needed to scale with confidence.”

— Venkat Cherukuri, Director of AI Solutions at PepsiCo

Better together

Crew Studio helps domain experts and engineering teams turn business processes into deployable agent workflows. Arize AX adds the evaluation and experimentation layer: traces, evaluations, labeling workflows, datasets, and experiments that help teams understand behavior and validate changes.

Together, the platforms support a continuous development workflow: build in Studio, deploy through CrewAI AMP, inspect and evaluate runs in Arize, test potential fixes, and redeploy with greater confidence.

The integration is built on OpenTelemetry standards, so teams retain control over their architecture. Arize works across agent frameworks and model providers, while Studio can export telemetry to compatible backends. The native integration gives joint customers a direct, supported path between the two platforms.

Learn more

Crew Studio announcement

Arize AX: OpenTelemetry and OpenInference semantic conventions

Arize AX docs

CrewAI documentation

The post Crew Studio launches with native Arize AX tracing and evaluation appeared first on Arize AI.