Releases·Tool Release
Automated agent evaluation with Amazon Bedrock AgentCore and GitHub Actions
An AWS guide shows how to build a CI/CD quality gate that deploys an agent on Amazon Bedrock AgentCore with role-based MCP tools, evaluates each change, and blocks PRs when evaluation scores drop. It gives practitioners a pattern for catching agent regressions before production.
0 engagement1 sourceTue, Sep 8, 2026, 04:23 PM
AI Overview
- AWS published a GitHub Actions CI/CD quality gate that deploys a Strands agent to Bedrock AgentCore runtime, evaluates it via the AgentCore Evaluate API, and blocks PRs on regression.
- A shared Cognito user pool serves both M2M (client_credentials) and user (authorization_code) flows; M2M tokens bypass role checks while user tokens carry custom:roles claims.
- The Evaluate API scores OpenTelemetry spans with LLM-as-a-judge by default (Lambda-based code evaluation optional); each call must contain spans from a single session or it throws ValidationException.
- AWS outlines three auth approaches: evaluate stored trace fixtures, cache a refresh token in Secrets Manager, or dual-grant M2M/user tokens on the MCP server; the reference implementation uses the third.
- The reference implementation uses CDK to deploy two AgentCore runtimes behind FastMCP middleware enforcing role-based tool access, with the full stack in the awslabs/agentcore-samples repository.
Related
Releases·Tool Release
AWS demos multi-agent supply chain system with explainability and custom evaluators on Amazon Bedrock AgentCore
AWS published a technical walkthrough of evaluating multi-agent systems using Amazon Bedrock AgentCore Evaluations with built-in and custom evaluators. The reference implementation is a supply chain decisioning system with an orchestrator and four specialized sub-agents built on Strands Agents SDK, showing how teams can measure accuracy, business validity, and explainability from development through production.
AWS published a technical walkthrough of evaluating multi-agent systems using Amazon Bedrock AgentCore Evaluations with built-in and custom evaluators. The reference implementation is a supply chain decisioning system with an orchestrator and four specialized sub-agents built on Strands Agents SDK, showing how teams can measure accuracy, business validity, and explainability from development through production.
0 engagement1 sourcerssMon, Oct 5, 2026, 03:50 PM
Products·building
A shared agentic platform for Wood Mackenzie, built on Amazon Bedrock AgentCore
An agentic platform AWS is building with Wood Mackenzie on Amazon Bedrock AgentCore, addressing the gap between agent prototypes and production deployments. It standardizes concurrency, session isolation, identity, state, scaling, and guardrails so the energy-research firm doesn't rebuild this layer per agent. LLM-based agents run on real traffic with monitoring to verify correct behavior.
An agentic platform AWS is building with Wood Mackenzie on Amazon Bedrock AgentCore, addressing the gap between agent prototypes and production deployments. It standardizes concurrency, session isolation, identity, state, scaling, and guardrails so the energy-research firm doesn't rebuild this layer per agent. LLM-based agents run on real traffic with monitoring to verify correct behavior.
0 engagement1 sourcerssThu, Sep 17, 2026, 03:41 PM
Products·launched
HEMA internal knowledge assistant on Amazon Bedrock AgentCore with MCP
HEMA, the Dutch retailer, built an internal knowledge layer so engineers can get instant answers instead of navigating disconnected wikis, service catalogs, and IT portals. It uses MCP and Amazon Bedrock AgentCore to serve its technology organization. The post is an AWS case study co-written with HEMA engineers.
HEMA, the Dutch retailer, built an internal knowledge layer so engineers can get instant answers instead of navigating disconnected wikis, service catalogs, and IT portals. It uses MCP and Amazon Bedrock AgentCore to serve its technology organization. The post is an AWS case study co-written with HEMA engineers.
0 engagement1 sourcerssWed, Sep 23, 2026, 06:41 PM
Social·Community
1,000 Agent Tasks in a Sandbox: What Breaks When LLMs Run Code — Kevin Orellana, AWS
Kevin Orellana, an engineer on Amazon Bedrock AgentCore, wrapped AgentCore's code sandbox and cloud browser in an MCP server and had a coding agent generate and run 1,000 tests across 10 domains, logging failures to SQLite. The talk, recorded at the AI Engineer World's Fair 2026, surfaces interface-layer bugs (image encoding, null bytes, unreadable files) and sandbox-layer limits (memory, file permissions, seccomp, eBPF). His takeaway: teams should fuzz-test their agent sandboxes remotely.
Kevin Orellana, an engineer on Amazon Bedrock AgentCore, wrapped AgentCore's code sandbox and cloud browser in an MCP server and had a coding agent generate and run 1,000 tests across 10 domains, logging failures to SQLite. The talk, recorded at the AI Engineer World's Fair 2026, surfaces interface-layer bugs (image encoding, null bytes, unreadable files) and sandbox-layer limits (memory, file permissions, seccomp, eBPF). His takeaway: teams should fuzz-test their agent sandboxes remotely.
6 engagement1 sourceyoutubeMon, Oct 12, 2026, 05:36 AM
Products·launched
AI agent evaluation framework for production using Strands and AgentCore
Motorway and AWS PACE built an AI-powered dealer stock search agent that lets dealers find vehicles using natural language queries. To ensure reliability in a high-stakes auction environment, they developed a production evaluation blueprint using Strands (a scenario-based testing framework) and AgentCore (an agent evaluation tool). The framework helps teams prove agent correctness before deployment.
Motorway and AWS PACE built an AI-powered dealer stock search agent that lets dealers find vehicles using natural language queries. To ensure reliability in a high-stakes auction environment, they developed a production evaluation blueprint using Strands (a scenario-based testing framework) and AgentCore (an agent evaluation tool). The framework helps teams prove agent correctness before deployment.
0 engagement1 sourcerssThu, Jul 23, 2026, 05:00 PM