Introduction

Originally Published August 2026, on ARCweb.com by Colin Masson, Director of Research for Industrial AI, ARC Advisory Group Inc.

Executive Takeaway

In a brownfield production facility and extended supply chain network, no single software vendor commands the entire technology stack. The rapid rollout of isolated AI assistants across ERP, EAM, TMS, and SCADA creates cognitive overload and dangerous cross-domain algorithmic collisions. To prevent operational paralysis, industrial enterprises must implement open inter-agent protocols—specifically Anthropic’s Model Context Protocol (MCP) and Agent-to-Agent (A2A) standards—to construct a governed master control plane that arbitrates conflicting operational, financial, and logistics objectives across Factory, Supply Chain, and Enterprise domains.

I. From Conversational Widgets to Agentic Swarms: The Copilot Tower of Babel

If you followed our opening installment of this series, you know we drew a firm line in the sand regarding what constitutes a true Industrial Copilot. As we dug deep into the market for ARC’s inaugural ARC Industrial Copilots MarketMap, we witnessed a fundamental (R)Evolution: the market is rapidly shedding the illusion of static, conversational sidecar chat widgets—those “Copilot-Washed” text boxes requiring operators to type 50-word prompts into a sidecar window—and evolving toward adaptive, context-aware Generative UIs (GenUI) driven by multi-agent reasoning engines.

Today’s advanced copilots don’t wait around for a human operator to type a text prompt. Grounded in living Industrial Knowledge Graphs (IKG) and real-time operational event streams, they actively sense anomalies, evaluate multi-constraint trade-offs behind the scenes, and present pre-validated decision trees directly at the point of work.

However, as these agent-driven copilots mature and multiply, they expose industrial leaders to an alarming new operational failure mode: The Copilot Tower of Babel.

In a modern brownfield enterprise, no single software vendor owns the entire end-to-end stack. As ERP giants, supply chain visibility networks, EAM platforms, and automation incumbents all launch their own autonomous digital workers, enterprise architects are suddenly managing dozens of uncoordinated “AI colleagues”:

• Enterprise ERP & EAM Giants (View C): SAP deploys Joule across 35+ enterprise applications; Infor introduces its Velocity Suite (anchored by the Infor Agentic Orchestrator and 100+ micro-vertical role agents running on Infor OS); IFS runs IFS.ai and IFS Loops for asset-intensive EAM/FSM; IBM integrates watsonx natively with Maximo Asset Management; Oracle embeds Fusion SCM & Maintenance Advisors.
• Supply Chain & Logistics Specialists (View B): FourKites deploys its autonomous digital workforce (Tracy for track-and-trace, Alan for scheduling, Sam for supplier collaboration, and YardWorks AI); Aera Technology orchestrates cross-enterprise decisions via the Aera Decision Cloud, Decision Ledgers, and prepackaged Aera Skills; Project44 runs its Ocean Exceptions Agent; o9 Solutions deploys APEX Agents on its Enterprise Knowledge Graph; Kinaxis powers concurrent planning via Maestro; Blue Yonder runs Orchestrator AI; Coupa deploys Navi Agents; Tacto automates direct CAD/BOM procurement.
• Factory, OT & Process Leaders (View A): Siemens embeds Industrial Copilot inside TIA Portal, NX, and Senseye; Schneider Electric deploys EcoStruxure Automation Expert Copilot; AVEVA operates CONNECT Assistant; Rockwell Automation deploys FactoryTalk Design Studio Copilot and Fiix CMMS; Honeywell injects Forge Intelligent Assistant into the DCS; SymphonyAI deploys 300+ IRIS Flows agents on its 7-trillion-data-point industrial model; Cognite runs Atlas AI on the Industrial Knowledge Graph; HighByte delivers first-mile Industrial DataOps and Industrial MCP server connectivity; XMPro provides composable, event-driven digital twin orchestration to connect edge telemetry with multi-agent decision swarms.

Without an overarching orchestration layer, this proliferation creates severe cognitive overload. When an operational exception occurs, which vendor copilot holds the authoritative truth? More dangerously, what happens when competing vendor agents issue conflicting recommendations based on isolated data views?

Consider a real-world, three-way value chain collision:

1. The OT / Factory Copilot: A predictive maintenance agent analyzing live bearing telemetry detects a high-frequency micro-friction anomaly on a primary resin extruder at Plant 2. Operating on an OT mandate to prevent mechanical failure, it recommends an immediate 30 percent throughput reduction or a 4-hour maintenance shutdown.
2. The IT / Enterprise ERP Copilot: An ERP procurement and financial agent reading quarter-end revenue targets and high-margin spot customer commitments recommends a 15 percent line speed acceleration to ensure billing cutoffs are met before midnight.
3. The SCM / Supply Chain Copilot: A logistics tracking agent detecting a 12-hour maritime port delay for raw polymer shipments recommends canceling local carrier pickups and rerouting inbound raw materials to Plant 4, completely blind to the fact that Plant 2 is about to run out of feedstock or that Plant 4 has zero warehouse storage capacity.

When these three conflicting recommendations arrive simultaneously in separate vendor sidecar chat boxes on different managers’ screens, who wins? In an uncoordinated environment, these conflicting instructions cause control loop hunting, stranded inventory, rapid equipment degradation, and operational paralysis.

II. MCP, A2A, and CESMII i3X Protocols as the Universal Grammar

To resolve this “Tower of Babel,” industrial pacesetters are rejecting single-vendor monopolies and implementing three foundational open integration standards:

1. Model Context Protocol (MCP) as the Universal Execution Grammar: Standardizes how a master enterprise copilot or workflow engine securely queries tools, databases, and sub-agents across third-party software stacks without custom N × M glue code. Middleware, DataOps, and composable twin platforms such as the  HighByte Industrial MCP Server, XMPro Event-Driven Orchestration, Microsoft Azure AI Foundry, Siemens Intelligence Center X, and Cognite Atlas AI utilize MCP to expose operational data pipelines as discoverable, server-side tools.
2. Agent-to-Agent (A2A) Inter-Agent Collaboration: Utilizes open inter-agent standards that allow specialized vendor copilots to communicate, discover capabilities, and negotiate trade-offs behind the scenes before presenting a unified recommendation to a human operator or planner.
3. CESMII i3X & Smart Manufacturing Profiles for Standardized Asset Semantics: Leverages the Clean Energy Smart Manufacturing Innovation Institute (CESMII) Interoperable Industrial Information Exchange (i3X) framework and open Smart Profiles to standardize first-mile machine information models and asset metadata. While MCP standardizes the dynamic tool execution grammar and A2A enables inter-agent trade-off negotiation, CESMII i3X Smart Profiles ensure that digital agents across Factory, Supply Chain, and Enterprise domains interpret machine states, sensor telemetry, and process variables with unified semantic fidelity—eliminating costly manual tag mapping and custom data model debt.

Real-World A2A Interoperability Across Extended Supply Networks

For a publicly announced, productized benchmark delivering this exact style of cross-domain value chain orchestration, look no further than Cognite’s joint Integrated Supply Chain offering with FourKites and Deloitte. As I explored following Cognite Impact in Oslo (see my detailed analysis: Oslo, Uncarpeted: Escaping the Desk for a Cyber-Physical Reality Check), unifying plant-floor operational data from Cognite Data Fusion® with FourKites’ real-time transportation telematics allows planners and plant managers to break down the walls between manufacturing (“Make”) and logistics (“Deliver”) in real time via open inter-agent protocols.

Other leaders across the Supply Chain Decision Intelligence arena—including Aera Technology, o9 Solutions, Kinaxis, and Blue Yonder—are singing from the exact same songbook.

Take Aera Technology as an exemplar of this agentic architecture. Operating via the Aera Decision Cloud, policy-driven Aera Skills, and the Aera Decision Ledger, Aera’s decision agents are designed to ingest real-time external transportation exception feeds from logistics visibility platforms via open A2A interfaces. When an unexpected ocean freight delay or port congestion event occurs, that signal isn’t dumped into a static spreadsheet or an unread email chain. Aera’s decision agents automatically evaluate the financial impact, check active inventory policies across the Decision Ledger, and simulate alternative re-routing scenarios—cross-referencing plant-floor production schedules before presenting pre-validated action cards to update S/4HANA order logs or execute a supplier expedite under human sign-off.

By allowing vendor-specialized agents to communicate natively across open boundaries, enterprises bridge the gap between logistics execution and corporate decision intelligence without accumulating custom N × M API debt.

III. Categorizing Domain Authority & the Master Control Plane

To safely govern a multi-copilot ecosystem, industrial buyers must define clear boundaries of Domain Authority while strictly enforcing human-in-the-loop governance:

• View A: Factory Copilot Authority: Operates over high-frequency OT telemetry, sub-second PLC/DCS code, physical interlocks, and safety envelopes (SIL/GxP). Factory Copilots hold primary domain authority for evaluating machine health and generating physics-bounded recommendations at the machine face, but strictly require human operator validation (via e-signatures or User-Sanctioned Setpoint Revision / USSR gates) before closing any physical control loop. Closed-loop autonomous execution remains the exclusive domain of Level 3/4 Autonomous Execution Agents (check out Greg Gorbach’s definition and upcoming companion MarketMap) operating within hard-coded physical interlocks.
• View B: Supply Chain Copilot Authority: Operates over telematics, yard management, carrier tracking, demand/supply trade-offs, and direct CAD/BOM procurement. Supply Chain Copilots optimize logistics flows and sourcing recommendations, but strictly require human planner validation before executing financial commitments or carrier re-routings.
• View C: Enterprise Copilot Authority: Operates over S/4HANA/EAM, P&L financial ledgers, cross-site corporate strategy, and compliance audit trails. Enterprise Copilots translate plant/SCM anomalies into financial insights and govern multi-agent security, but require human sign-off for transactional execution and cannot override physical OT safety gates.

IV. Key Takeaways & Executive Diagnostic Framework

When auditing software vendor RFIs for multi-vendor copilot deployments, use these four essential diagnostic inquiries to evaluate cross-domain orchestration readiness:

• Cross-Domain Collision Arbitration: How does your copilot discover, communicate with, and negotiate trade-offs with third-party AI agents in other systems (e.g., how does your ERP copilot negotiate with our APM or TMS copilot)?
• Open Protocol & Semantic Standard Compliance (MCP, A2A & CESMII i3X): Does your solution natively expose its capabilities as a Model Context Protocol (MCP) Server, leverage CESMII i3X Smart Profiles for standardized asset semantics, and support open Agent-to-Agent (A2A) interfaces to eliminate N × M custom API and data modeling debt?
• Domain Authority Envelopes: How does your architecture enforce hard domain authority boundaries to prevent top-floor financial copilots from writing ungated setpoints directly to PLCs or historians?
• Multi-System Context Federation: Does your copilot query a single, zero-copy Industrial Knowledge Graph or Unified Namespace to resolve cross-functional context (Source-Make-Deliver-Maintain), or does it rely on isolated data copy pipelines?

About the author

C, Masson (25/08/2026) (2) The “Copilot Tower of Babel”: Orchestrating Multi-Vendor Copilots Across the Extended Industrial Enterprise | LinkedIn