Agent Platform Traffic Splitting for A/B Testing

PMLE By islever Editorial Last updated

Scaling online model serving.

Short answer

B Deploy the new model to the Agent Platform endpoint where the current model is deployed and configure the endpoint’s traffic split to allocate 10% of user requests to the new model.

Question

You are responsible for managing the recommendation engine on an ecommerce website. The current model is deployed on a public Agent Platform endpoint and is actively serving production traffic. You have developed an updated version of the model and registered the model in Model Registry. You need to perform an A/B test to compare the new model’s performance to the current model.
You plan to implement a traffic split, routing 10% of all incoming production requests to the new model and sending the remaining 90% to the current model. You want this transition to be transparent, and require no modifications to the client application that calls the endpoint. What should you do?
  1. Deploy the new model to a serverless Cloud Run service. Implement logic in the client application that routes 10% of user requests to the Cloud Run service and the rest to the original endpoint.
  2. Deploy the new model to the Agent Platform endpoint where the current model is deployed. Configure the endpoint’s traffic split to allocate 10% of user requests to the new model. Correct Answer
  3. Create two separate Agent Platform Pipelines, one for each model version. Use Agent Platform Experiments to run both pipelines simultaneously. Configure the experiment to send 10% of user requests to the new model’s pipeline and the rest to the production model’s pipeline.
  4. Deploy the new model to a new Agent Platform endpoint. Configure Cloud Load Balancing to send 10% of user requests to the new endpoint and the rest to the original endpoint.

Key Takeaways

The trap is thinking you need external load balancing or client changes; the Agent Platform handles traffic splits natively at the endpoint level.

Learn how to implement transparent A/B testing for Gemini Enterprise Agent Platform models using native traffic splitting features without client-side changes.

Common mistake: Choosing D (Cloud Load Balancing) is a common mistake because it seems like standard practice for traffic splitting, but it ignores the native, simpler capability provided by the Agent Platform's endpoint configuration.

Expert Analysis

The scenario requires performing an A/B test on a recommendation engine deployed via the Gemini Enterprise Agent Platform (referred to as Agent Platform in the options). The key constraints are that the current model is live, an updated version is registered in the Model Registry, and the transition must be transparent with no modifications to the client application.

To achieve this, the routing logic must reside within the serving infrastructure itself, rather than requiring the client to make decisions or external load balancers to manage traffic distribution. In the Gemini Enterprise Agent Platform ecosystem, endpoints support traffic splitting configurations directly. By deploying the new model version to the same endpoint configuration as the existing one (or associating it with the active endpoint), you can define a split ratio (e.g., 10% to the new version, 90% to the current version). This allows the platform to handle the routing internally. Clients continue to send requests to the single, unchanged endpoint URL, satisfying the transparency requirement.

Option B correctly describes this mechanism: deploying the new model to the endpoint and configuring the traffic split. Option A fails because it requires modifying the client application to route traffic. Option D introduces Cloud Load Balancing, which adds complexity and potentially changes the endpoint structure or requires client awareness if not perfectly abstracted, but more importantly, the Agent Platform provides native traffic splitting capabilities for this exact use case without needing external LBs. Option C misuses Pipelines and Experiments; these tools are for development and testing workflows, not for serving live production traffic splits.

Exam Strategy

When dealing with Google Cloud AI services, always look for native management capabilities before assuming external infrastructure is needed. For Agent Platform questions involving model updates or A/B testing, check if the endpoint supports traffic splitting or versioning natively. This often eliminates the need for complex client-side logic or external load balancers, leading to simpler, more maintainable solutions.

Frequently Asked Questions

Can I use Cloud Load Balancing for Agent Platform A/B testing?

While technically possible, it is unnecessary and complex. The Agent Platform provides native traffic splitting at the endpoint level, which is the intended and simpler method for transparent A/B testing.

Do I need to change the client application for A/B testing?

No. By using the Agent Platform's native traffic splitting feature, the routing happens server-side. The client continues to call the same endpoint URL, making the test completely transparent.

More PMLE FAQ →

Community Votes

B
100%

100% of anonymous learners picked answer B. Votes are pick records left by other test-takers — they are not the verified answer.

Discussion 2 contributions

Compiled from public community threads and reader submissions, lightly edited for clarity. Handles and vote tallies are shown as recorded, and comments reflect the views of individual contributors.

From public community threads 2

  1. taylz876 👍 1 Answered B
    Repeated Q234, Q266 and Q323 (the same traffic-splitting scenario) Why B is right • Same endpoint means no change for clients: callers keep using the same address, so the transition is transparent and the client application doesn't change. • Traffic splitting is built in: the endpoint sends 10% of requests to the new model and 90% to the current one, and it's a config change, not a code change. Why the others are wrong • A: Routing logic in the client application is a client modification, which the question rules out. • C: Experiments tracks runs and doesn't route live traffic. • D: A new endpoint behind Cloud Load Balancing adds an extra layer to build and maintain, when the endpoint already splits traffic.
  2. n2183712847 👍 1 Answered B
    Correct answer: B. Deploying both model versions to the same Agent Platform endpoint and configuring a 10/90 traffic split provides transparent A/B testing without requiring any client application changes; A and D require additional routing infrastructure or client changes, while C misuses Pipelines and Experiments, which are not designed for routing live production traffic.

Reader comments & corrections

No reader corrections yet — spotted an error or have a note? Share it below.

Log in to comment, report an error, or add a note about this question.

Monitoring ML Model Performance with Precision and Recall Sampling

S-grade

Optimizing ML Model Recall for Compliance Risk

S-grade

How to Minimize Development Time for Weekly TensorFlow Retraining in Vertex AI?

S-grade

Building CLV Models with BigQuery ML in Vertex AI Workbench

S-grade

Practice All PMLE Questions

Access 65 questions with complete answers and detailed explanations.

View Full PMLE Practice Test →

← Back to PMLE Study Guide