Weekly signal

This week (May 11–19, 2026) the manufacturing sector moved from experimental agent pilots toward productized, industry-focused agent stacks: industrial-scale agent platforms from Emerson/AspenTech and Rescale; factory-grade agentic operating-system functionality from Hermes Reply; open-source agent training frameworks (Orchard) and field research highlighting an adoption/verification gap and new cyber risks. The dominant thread: vendors are packaging agentic workflows that tie LLM decision-making to OT/CAE data and simulation — but academic and practitioner research warns that verification, governance, OT readiness, and attacker economics remain critical blockers.

What changed

1. Emerson (AspenTech) launched AVA, an "agentic, domain-aware" industrial AI platform that embeds first‑principles models and OT context to deliver agentic decision-support across operations. AVA is positioned as a production-ready advisor layer for process and discrete industries.

2. Rescale announced "agentic digital engineering": simulation-native AI agents that automate CAE workflows (input validation, troubleshooting, surrogate creation) to compress R&D cycles and push agentic workflows into product development and upstream manufacturing decisions. Customers cited include Daikin and other large industrials.

3. Hermes Reply (Reply Group) introduced Brick Cognitive, an agentic operating system integrated with MES/MOM functionality that exposes prebuilt manufacturing advisors (quality investigation, KPI advisor, production flow advisor) — a vendor move to productize agents inside shop‑floor stacks.

4. Research and tooling advanced: Microsoft Research / Orchard published an open-source agentic modeling framework (training/recipes for task-specific agents), while a separate arXiv study of industrial adopters found most firms at early maturity and highlighted a capability→deployment verification gap (lack of production-grade verification & context grounding).

5. Infrastructure and risk signals: NVIDIA updated Omniverse on DGX Cloud (May 15) — reinforcing simulation and digital-twin infrastructure for agent testing — while security research flagged that agentic AI compresses attack lifecycles and elevates OT/CI risk, making immediate defensive changes necessary.

What to do with it

• If you run factories: map 1–2 high-value, low-regret workflows (root-cause analysis, maintenance triage, CAE setup) and pilot agents with strong data grounding and human-in-the-loop gates. Start in simulation (Omniverse/Rescale) before production.
• For engineering teams: evaluate Orchard and other open recipes to prototype domain-specific agents, but require traceable decision logs and sandboxed tool calls. Expect to instrument verification tests as part of CI/CD for agents.
• For security and ops: treat agentic deployments as new attack surfaces — harden identity, telemetry, CI/CD, and agent governance now. Add agent-execution monitoring and recovery playbooks.
• For leaders: adopt an Agent Development Lifecycle (governance, testing, metricization) and budget a verification and OT-data readiness workstream before scaling.