Azure AI Foundry April Updates: New Capabilities for Enterprise Agents
Breaking down the latest Azure AI Foundry features including enhanced prompt optimization, distributed tracing, and improved RBAC.
What's New in April 2026
Microsoft has released significant enhancements to Azure AI Foundry designed to accelerate enterprise agent deployment and optimization.
1. Enhanced Prompt Optimizer
The new Prompt Optimization workflow automates iterative improvements to agent instructions:
Key Features:
- Automated performance benchmarking against test datasets
- Multi-variant A/B testing for prompt variations
- Quality metrics tracking (accuracy, latency, cost)
- Integration with Foundry batch eval system
Usage Example:
const optimizer = foundry.createPromptOptimizer({
agent: agentId,
testDataset: "dataset-123",
metrics: ["accuracy", "latency", "cost"],
iterations: 10
});
const optimizedPrompt = await optimizer.optimize();
const improvement = optimizedPrompt.metrics.accuracy -
baselinePrompt.metrics.accuracy; // Track improvementExpected Improvements:
- 15-25% accuracy gain on typical workflows
- 8-12% latency reduction
- 5-10% cost reduction
2. Distributed Tracing & Observability
Full end-to-end visibility into agent execution:
What You Get:
- Trace every tool call, LLM invocation, and data transformation
- Latency breakdown by component (LLM, tools, logic)
- Custom spans for application logic
- Integration with Application Insights
Trace Example:
{
"traceId": "trace-abc123",
"spans": [
{
"name": "llm-completion",
"duration": 1200,
"model": "gpt-4",
"tokens": { "input": 450, "output": 120 }
},
{
"name": "database-query",
"duration": 340,
"query": "SELECT * FROM customers WHERE..."
},
{
"name": "service-bus-publish",
"duration": 95,
"messageCount": 1
}
]
}3. Improved RBAC & Multi-Tenant Support
Enterprise-grade access control:
New Capabilities:
- Fine-grained roles for agent components (reader, operator, deployer)
- Cross-subscription resource access
- Service principal support with managed identities
- Audit logging for all permission changes
Example Role Configuration:
resource roleAssignment 'Microsoft.Authorization/roleAssignments@2022-04-01' = {
name: guid(resourceGroup().id, agentOperatorServicePrincipal.id)
scope: foundryProject
properties: {
principalId: agentOperatorServicePrincipal.id
roleDefinitionId: subscriptionResourceId(
'Microsoft.Authorization/roleDefinitions',
'a8fef144-21e9-436c-a6c9-8b3ab9c93062' // Foundry Agent Operator
)
}
}4. Cost Control Features
Manage agent spending at scale:
Features:
- Per-agent token budgets with warnings and limits
- Model/provider cost tracking
- Cost forecasting based on historical usage
- Integration with Azure Cost Management
Configuration:
const agent = foundry.createAgent({
name: "customer-support",
costBudget: {
monthlyLimit: 10000, // USD
warningThreshold: 8000,
alertEmail: "ops@company.com"
}
});5. Enhanced Dataset Management
Better support for evaluation and training data:
New Tools:
- Import evaluation traces directly into datasets
- Version control for dataset updates
- Data quality validation rules
- Integration with Azure Data Explorer
6. Improved Container Deployment
Simpler path to production:
What's New:
- One-command push to Azure Container Registry
- Automatic scaling policies based on load
- Health probes and readiness checks
- Graceful shutdown handling
Deployment:
azd deploy --agent-id <agent-id> --registry myacrMigration Guide: v1 → v2 API
If you're upgrading from Azure AI Foundry v1:
Breaking Changes
| v1 API | v2 API | Migration Path |
|---|---|---|
| `agent.invoke()` | `agent.run()` | Rename method calls |
| `config.llm` | `config.models[]` | Update config structure |
| String trace IDs | UUID format | System handles automatically |
Migration Example
Before (v1):
const result = await agent.invoke({
input: "Hello",
config: { llm: "gpt-4" }
});After (v2):
const result = await agent.run({
input: "Hello",
config: { models: [{ provider: "azure-openai", id: "gpt-4" }] }
});Performance Benchmarks
Testing on 100 common enterprise agent patterns shows:
| Metric | v1 | v2 | Improvement |
|---|---|---|---|
| Agent startup | 850ms | 320ms | 62% faster |
| Tool invocation | 450ms | 380ms | 16% faster |
| Token throughput | 18K/s | 24K/s | 33% faster |
| Memory usage | 450MB | 280MB | 38% lower |
| Dashboard latency | 2.3s | 0.8s | 65% faster |
Upgrade Timeline
Critical: v1 support ends June 30, 2026
- March 15–May 31: Parallel support for v1 and v2
- June 1 onwards: v2 only; automatic migration available
- July 1 onwards: v1 APIs deprecated
Recommendations
1. Start now: Review the v2 API documentation 2. Create v2 agent: Deploy new agents on v2 architecture 3. Migrate gradually: Move existing agents to v2 (most migrations take <4 hours) 4. Leverage new features: Use Prompt Optimizer and distributed tracing for production improvements
Resources
- Docs: aka.ms/foundry-v2-guide
- Demo: YouTube: Foundry v2 Feature Walkthrough
- Support: foundry-team@microsoft.com
---
What features are you most excited about? Share your feedback in the comments below.