PwC’s AI agents are now your consultants — whether you’re ready or not. PwC launches PwC One, letting clients run autonomous agents for consulting tasks while professionals review outputs. If you build agentic systems, this is a live example of taking agents from lab to billable work — plan for human-in-loop review, SLAs, and orchestration (Principles 03 & 09).

Ai2 releases MolmoWeb, an open-weight visual web agent using screenshots (4B & 8B). AI2 ships MolmoWeb, a screenshot-driven visual web agent in open weights that automates navigation and tasks on real websites. An open visual agent lowers the bar for building agents that interact with real UIs and supports reproducible context engineering and toolchains for web automation (Principles 06 & 11).

What is DeerFlow 2.0 — what enterprises should know about this powerful local AI agent orchestrator. ByteDance open-sources DeerFlow 2.0, a Docker-sandboxed, model-agnostic orchestrator for secure, long-horizon local AI workflows. Local, sandboxed orchestration shifts deployment tradeoffs toward enterprise control, observability, and safer execution — evaluate sandboxing, connectors, and long-run state handling (Principles 07 & 09).

NanoClaw Adopts OneCLI Agent Vault. NanoClaw routes agent requests through OneCLI’s Agent Vault so agents never hold raw API keys and operate under enforceable policies. Credential proxies and policy-enforced secret handling are foundational for production agents — bake token brokers, audit logs, and gate controls into your security posture (Principles 10 & 15).

New JetBrains platform manages AI coding agents. JetBrains Central centralizes control, execution, and governance for team-scale AI coding agents across IDEs, pipelines, and tools. Tooling that manages coding agents highlights the need for governance, traceability, and artifact-level proof so teams can ship agent-produced code safely and integrate it into developer workflows (Principles 09, 10 & 06).