{"slug": "building-production-ready-ai-agents-with-atlas-agent-engine-and-arize-ax", "title": "Building production-ready AI agents with Atlas Agent Engine and Arize AX", "summary": "Arize AI is a launch partner for MongoDB's Atlas Agent Engine, a new platform for building, deploying, and governing AI agents in production, with Arize AX adding trace-level observability and evaluation. The integration lets Agent Engine agents emit OpenTelemetry traces that Arize AX ingests natively via the OpenTelemetry GenAI semantic conventions and OpenInference, so any agent runtime emitting OTLP on Agent Engine can send traces without custom instrumentation code. Arize AX reconstructs executed agent paths through agent graphs, supports online evaluations against production traffic, routes failures to annotation queues, and offers the Signal agentic system that identifies failures and suggests fixes with a proposed pull request.", "body_md": "Arize AI is a launch partner for MongoDB’s Atlas Agent Engine, a new platform for building, deploying, and governing AI agents in production.\n\nMongoDB Atlas is the platform where agents act on real-time data. Agent Engine provides the governed foundation for operating those agents, while Arize AX adds trace-level observability and evaluation to help teams understand, test, and improve agent behavior.\n\nTogether, Agent Engine and Arize AX connect the platform where agents act on enterprise data with the workflow teams need to inspect runs, evaluate outcomes and execution paths, identify failures, and redeploy with evidence.\n\nIn this article, we’ll explain how the integration works, what telemetry flows into Arize AX, and how to configure the connection.\n\n## **What is Atlas Agent Engine?**\n\nAtlas Agent Engine provides a governed foundation for building, deploying, and operating AI agents in production It is designed to work across models and frameworks while allowing enterprise teams to use their existing authentication and security tools.\n\nThat means organizations can build on the technologies they already trust, preserve flexibility, and continue evolving their technology stack as the market moves. Without rebuilding their agent infrastructure.\n\nMongoDB is launching it with partners helping enterprises move from agent prototypes to production. Arize AI is a launch partner, bringing trace-level observability and evaluation to help teams understand each agent run, measure outcomes and execution paths, and improve agents before redeploying.\n\nTogether, Agent Engine and Arize help enterprises build, operate, and improve production-ready agents with greater confidence.\n\n## **Why production agents need trace-level evidence**\n\nAn agent can fail even when its final response looks plausible. It may retrieve the wrong document, call the wrong tool, repeat a step, or use an expensive model where a smaller one can work just as well. Debugging these failures requires telemetry from the full execution path, and not just the final output.\n\nAgent Engine gives agents access to governed operational data and a place to build and run them. Arize AX ingests trace data from those runs so teams can inspect model calls, retrieval steps, tool calls, latency, and token usage at the span level. [Agent graphs](https://arize.com/docs/ax/instrument/agent-trajectory) in Arize AX reconstruct the path that was actually executed, making it easier to identify where behavior diverged from expectations.\n\nTraces show what happened while evaluations determine whether the behavior met your criteria. Teams can run [online evaluations](https://arize.com/resources/online-llm-evaluations/) against production traffic, score both the final output and execution path, and route failures to [annotation queues](https://arize.com/docs/ax/evaluate/human-review) for review. Reviewed examples can become datasets for [experiments](https://arize.com/docs/ax/evaluate/run-evals-on-experiments), allowing teams to compare a proposed change before redeploying. Arize AX also offers a built-in agentic system called [Signal](https://arize.com/blog/debug-production-ai-agents-with-signal-tutorial/) that automatically identifies failures and issues before suggesting fixes with a proposed pull request.\n\n## **Built on OpenTelemetry standards**\n\nArize AX natively ingests traces that follow the [OpenTelemetry GenAI semantic conventions](https://arize.com/docs/ax/concepts/otel-openinference/semantic-conventions), alongside [OpenInference](https://arize.com/docs/ax/concepts/otel-openinference/overview). Any agent runtime that emits OTLP on Agent Engine can send traces to Arize without custom instrumentation code. Because the integration rests on open standards, teams keep their flexibility across frameworks, models, and observability backends.\n\n## **How to connect Atlas Agent Engine to Arize AX**\n\n### **Before you open Trace Export**\n\n1. In Arize, get your Space ID, API key, and model / project ID (model_id). (In Arize AX, go to Settings, then API Keys. Copy your Space ID and create or copy an API key.)\n2. In Agent Engine, open the project → Secrets.\n3. Create a project secret whose value is only the Arize API key (no `api_key:` prefix). Remember the secret name (example:`ARIZE_API_KEY` ).\n\n### **Open the Trace Export drawer**\n\n1. Open the same project.\n2. Go to Traces.\n3. Click the Export chip in the header (OFF, or later Export (Arize) / ON).\n4. The Trace Export drawer opens.\n\n### **Configure Arize**\n\n1. Turn on Enable trace export.\n2. Under Destination:\n  - Preset → Arize\n  - (optional: click Setup guide for Arize’s OTEL docs)\n  - Confirm OTLP endpoint is set to: `https://otlp.arize.com/v1/traces`\n  - Protocol → HTTP/protobuf\n  - `space_id` → your Arize Space ID\n  - `model_id` → your Arize model/project ID (routes spans on Arize’s side)\n  - Auth secret reference → the secret name from step 3 above (e.g. `ARIZE_API_KEY` )\n  - Auth header name → `api_key` (Arize preset prefills this)\n3. Under Delivery:\n  - Egress mode → Agent Engine + customer mirror (or Platform + customer mirror if branding flag is off). Required while export is on.\n  - Content mode → Full content (only option that works today; sends prompts/completions/tool payloads)\n  - Sampling → Always on\n  - Leave Skip TLS verification unchecked\n4. Click Save settings.\n5. Drawer closes; toast should say Export is on. Chip should show Export (Arize) / ON.\n\n### **Check it worked**\n\n1. Run an agent invocation in that project.\n2. In Arize, confirm spans appear under that Space / model.\n3. Chip on Traces should stay ON.\n\n## **Use Arize AX to build better agents on Atlas Agent Engine**\n\nMoving agents from pilot into production requires more than model performance. Enterprises need a platform that connects agents to real-time data, applies governance by default, and preserves flexibility as models and frameworks evolve. Atlas Agent Engine provides that foundation, while partners like Arize help teams understand, evaluate, and improve agent behavior in production.\n\n*Erica Volini, Chief Customer Officer, MongoDB*\n\nAgents are moving from demos into production workflows, where reliability depends on both the context they act on and the ability to understand what happened at runtime. MongoDB gives agents access to the data they need, while Arize traces each run and supports agent evaluation across both the outcome and the path taken. We’re excited to be a launch partner for Atlas Agent Engine, helping enterprises ship agents they can understand, evaluate, and improve with confidence.\n\n*Jason Lopatecki, Co-Founder and CEO, Arize AI*\n\nMongoDB provides the intelligent data and agent platform: the context agents act on, and the place to build, deploy, and govern them. Arize AX adds the evaluation and experimentation layer: traces, online evals, labeling workflows, datasets, and experiments that show how agents behave and whether a change made them better.\n\nTogether, the workflow is continuous. Build and deploy on Agent Engine, inspect and evaluate runs in Arize, test fixes against real production data, and redeploy with evidence instead of hope.\n\nArize works across agent frameworks, model providers, and data stores. Agent Engine is neutral across models and frameworks. Neither locks you in, and the integration between them gives joint customers a direct, supported path from the first run.", "url": "https://wpnews.pro/news/building-production-ready-ai-agents-with-atlas-agent-engine-and-arize-ax", "canonical_source": "https://arize.com/blog/using-arize-ax-with-mongodb-atlas-agent-engine/", "published_at": "2026-09-30 15:00:55+00:00", "updated_at": "2026-09-30 15:20:57.502545+00:00", "lang": "en", "topics": ["ai-agents", "artificial-intelligence", "mlops", "ai-tools", "developer-tools"], "entities": ["Arize AI", "MongoDB", "Atlas Agent Engine", "Arize AX", "OpenTelemetry", "OpenInference", "Signal"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/building-production-ready-ai-agents-with-atlas-agent-engine-and-arize-ax", "markdown": "https://wpnews.pro/news/building-production-ready-ai-agents-with-atlas-agent-engine-and-arize-ax.md", "text": "https://wpnews.pro/news/building-production-ready-ai-agents-with-atlas-agent-engine-and-arize-ax.txt", "jsonld": "https://wpnews.pro/news/building-production-ready-ai-agents-with-atlas-agent-engine-and-arize-ax.jsonld"}}