AWS expands serverless fine-tuning, adds agent analytics and automated policy checks

Amazon Web Services announced on August 3, 2026 that Amazon SageMaker AI serverless model customization now supports full fine-tuning for over 25 open-source models, including Llama, Gemma, Nemotron, Qwen, and gpt-oss families. The update expands beyond parameter-efficient methods like LoRA, which update only a subset of weights.

Full fine-tuning updates all model parameters, enabling deeper domain adaptation for specialized reasoning, complex output formats, and internalization of proprietary knowledge. The serverless option handles infrastructure provisioning and training orchestration, with pay-per-use pricing. It is available in US East (N. Virginia), US West (Oregon), Asia Pacific (Tokyo), and Europe (Ireland).

Serverless fine-tuning meets agent observability

Separately, Y Combinator-backed startup Armature (YC P26) launched product analytics for agent sessions on the Model Context Protocol (MCP). The company, founded by Theodore and Louis, provides an SDK that wraps an MCP in three lines of code, available in TypeScript, Python, and Go.

• Reconstructs the full session behind MCP tool calls, including the user's original request and the agent's reasoning.
• Ranks the most popular use cases for each MCP based on session clustering.
• Identifies the most frequent issues that user agents encounter, helping developers fix problems faster.
• Dashboard displays reconstructed conversations as if reading the real interaction inside Claude or ChatGPT.

Automated reasoning for policy refinement

Amazon Bedrock also introduced automatic Automated Reasoning policy refinement. The refinement engine diagnoses failing tests and proposes formal-logic fixes for rule issues and language issues. Every proposed change requires user approval before it takes effect. The feature is accessible via both API and the Bedrock console.

These three developments lower barriers for fine-tuning and deploying AI models. SageMaker's full fine-tuning gives teams deeper customization without managing infrastructure. Armature's analytics bring visibility into agent behavior, a growing need as MCP adoption increases. Bedrock's automated reasoning helps enforce safety policies with less manual debugging. Together, they point to a maturing ecosystem where customization, observability, and governance are becoming automated and serverless.