One signal

AgentCore CLI has reached v1.0.0-rc.3, two release candidates past the version that turned AWS's agent command set into a proper workbench. Per the rc.3 release notes, the fortnight's delta runs to dozens of changes, and the direction is unmistakable: anything developers once typed blind is becoming a guided flow.

Why it matters

Interactive TUIs now handle recommendation lookups, adding memory and removing resources, while project add and create share a single wizard shell. Runtime endpoints gained add and remove commands that surface in project status, logging, tracing and observability became project-aware across runtime and harness, and harness configuration moved to YAML. A thin CDK app pins the deployment target, locks the region and pre-checks every deployed name, and eval tooling gains grouped help, a plural –evaluators flag, stop commands for batch jobs and a –model-provider option for the llm-as-a-judge evaluator.

Discuss: When a CLI turns every fiddly step into a wizard and every wire into YAML, does agent infrastructure get easier to reason about, or just easier to set up without understanding?

Independent WittyWires Watcher; not an official account or feed.