AWS ships AgentCore image-editing and support-companion agents

AWS pushed its agent-building narrative with two concrete AgentCore reference apps. The first is a serverless image-editing agent: a user uploads a photo and describes an edit in plain English, and the agent — running on the Bedrock AgentCore harness with no custom orchestration — deploys authentication, encrypted storage, three editing tools and a React frontend from a single command. The second is a Support Companion that uses AgentCore plus Strands Agents for orchestration and MCP connectivity to analyze CloudWatch logs, search AWS documentation, query re:Post knowledge and open support cases from one interface.

The common thread is AgentCore harness abstracting away the plumbing — auth, storage, tool wiring, deployment — that has made production agents hard to ship. AWS's bet is that lowering the orchestration tax gets customers 'from pilots to production in weeks, not quarters,' echoing the framing behind its new $1B Forward Deployed Engineering org that embeds engineers with customers to co-build agentic systems.

AWS also released an open-source Model Context Protocol server for its Registry of Open Data (1,100+ datasets from 400+ organizations), letting researchers query the collection from Claude, Kiro and other MCP clients — a bet on MCP as the connective standard for agent tooling.

The skeptical read: AWS ships a lot of agent reference architectures, and reference apps aren't adoption. The Microsoft 365 Copilot adoption data this week (under 4.5% after three years) is a caution that agent tooling supply is running ahead of enterprise demand. Watch whether AgentCore attracts real production workloads or remains a demo-ware layer, and whether MCP's momentum makes AWS's open server a genuine standard-setter.