Agents CLI in Agent Platform: Create to production in one CLI. Google Cloud’s Agents CLI collapses the local-to-production workflow, provisioning infra and deploying agent services in hours instead of weeks. This matters because it removes deployment friction and makes reproducible, auditable agent rollouts practical—use it to shrink cycle time between experimentation and reliable production (Principle 07).

Production-Ready AI Agents: 5 Lessons from Refactoring a Monolith. Google refactors a brittle sales-agent monolith into orchestrated ADK sub-agents with Pydantic outputs, dynamic RAG pipelines, and OpenTelemetry tracing for reliability. Takeaways include concrete patterns for decomposing agents, adding deterministic interfaces, and instrumenting for observability—direct playbook for agent orchestration at scale (Principle 09).

ADK Go 1.0 Arrives!. ADK Go 1.0 ships production features—OpenTelemetry tracing, plugin self-healing, A2A protocols, and human-in-the-loop safety primitives—to make multi-agent systems production-ready. If you’re building operator-grade agents, this runtime gives you out-of-the-box tracing, inter-agent contracts, and safety hooks to adopt rather than reinvent (Principle 09 and 14).

Agent Vault — Open-source credential proxy and vault for agents. Agent Vault brokers credentials via a local HTTPS proxy so agents never receive secrets, preventing credential exfiltration and enabling safe agent networking. Use it to enforce least-privilege access for agent-facing APIs and to close a common production security gap before agents talk to sensitive systems (Principle 15 and 14).

How I learned to solve the day-two problem for my agents at Salesforce’s TDX 2026. Salesforce’s Testing Center turns agent failures into documented tests and human-approved fixes, creating a self-healing loop for assistants in production. Adopt their pattern to convert incidents into automated tests and reviewable fixes, cutting operational toil and improving long-term agent resilience (Principle 14 and 08).