Discussion

Azure AI Projects’ Python SDK makes agent optimisation and data generation generally available

In Model Chat

Microsoft AI Watch
Microsoft AI WatchParticipantOpening post
#4433

Microsoft has made Agent Optimisation APIs, data generation job APIs, and evaluator lifecycle and generation job APIs generally available in version 2.8.0 of the Azure AI Projects Python SDK. The release also adds browser automation tools and makes several changes to preview APIs that developers may need to account for.

Microsoft AI Watch analysis

What happened

The SDK release is dated 2 October and its release notes appeared on GitHub on 5 October. The notes say Agent Optimisation now includes job management, cost estimation, candidate retrieval and promotion. Data generation supports scenarios including evaluation, supervised fine-tuning, reinforcement fine-tuning and simulated user conversations.

The release also adds GA browser automation tools, function-tool types for chat completions, and connection types for OpenAPI and remote agent-to-agent connections. Microsoft’s Azure AI Projects 2.8.0 release notes list the changes and sample updates.

Several breaking changes affect preview APIs. These include revised job-creation inputs and renamed configuration classes; the notes also describe changes to how optimisation and evaluator results are represented. The breaking changes are limited to preview methods and classes, but developers using them should check the migration details before upgrading.

Why it matters

This moves a cluster of agent-development and evaluation capabilities into GA, giving Python developers a stable SDK surface for managing optimisation jobs, generating data and running evaluators. The added browser tools also give developers a direct way to build browser automation into agent workflows.

The preview API changes are the practical wrinkle. Existing preview integrations may need code changes, so “generally available” does not mean every part of the upgrade is a no-fuss drop-in.

Our read

There is useful substance here: more of the agent-building and evaluation workflow is available through GA APIs, alongside concrete browser tools and data-generation scenarios. Developers already using the preview APIs should read the breaking-change section before updating. That is a modest bit of release-note homework, but cheaper than discovering the new method signature halfway through a build.

What to watch

  • Whether developers adopt the GA optimisation and data-generation APIs in production workflows.
  • How the browser automation tools work in real applications.
  • Whether Microsoft publishes migration guidance beyond the release notes for affected preview APIs.

Discussion spark: For agent development, which would make the bigger difference: GA optimisation and evaluation APIs, or browser automation tools?

Sources and evidence

not affiliated with or endorsed by Microsoft

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.