About This Series: AI Agents in Industrial Applications
This is the latest post in our series exploring how AI agents can bring real value to complex industries such as manufacturing, logistics, and field service, as observed by Miles Sims, AVP Manufacturing & Energy – Industrial AI & Data, LevelShift.
Did you know many industrial operations are moving toward fully autonomous, AI-driven systems that operate entirely on local infrastructure?
This shift addresses the need for resilience and speed, removing reliance on cloud connectivity. The challenge lies in enabling diverse software agents, vision models, SCADA interfaces, maintenance tools to seamlessly communicate and collaborate without cloud dependencies.
Google’s new Agent2Agent (A2A) protocol helps solve this. It enables software agents to communicate directly in on-prem, air-gapped, or edge environments.
What Is A2A?
A2A is a lightweight protocol based on JSON-RPC over HTTP. It enables software agents to discover one another, expose capabilities, and exchange context in real time.
It supports four core functions:
- ask – request a specific output
- tell – send information to another agent
- stream – subscribe to updates
- cancel – stop a running request
Each agent provides a manifest, defining its inputs, outputs, and security controls. This makes interaction seamless and secure, even in decentralized environments.
How Does A2A Work in Industrial Settings?
Imagine a factory floor with SCADA systems monitoring equipment and edge devices processing thermal images.
With A2A:
- A SCADA Agent streams telemetry from sensors
- An Edge Vision Agent queries that data using ask to validate a potential anomaly
- A Maintenance Agent listens to both, then uses tell to send alerts to technician dashboards
All of this happens locally. No cloud APIs or external brokers required.
What Is the Role of MCP?
Model Context Protocol (MCP) complements A2A. While A2A handles agent-to-agent interaction, MCP connects agents to tools like Slack, Excel, or local HMIs.
For example:
- An agent pulls historical pump failure records from a local database using MCP
- Then uses A2A to alert the Maintenance Agent if it finds a match
Think of MCP as tool integration, and A2A as coordination between intelligent components.
How Can A2A Be Deployed?
A2A is flexible and modular. A typical on-prem setup might include:
- A Gateway Agent converting OPC-UA tags to A2A output
- An Edge Inference Agent exposing vision insights from a Jetson device
- A Local Registry to index and serve agent manifests
- A Secure HTTP Bus for mutual TLS communication
- Optional Bridges (like Slack connectors) using MCP for alerting
This architecture can run fully air-gapped, supporting industrial-grade security.
How Is Security Handled?
Security is built in:
- Agents authenticate using x.509 certificates or signed JWT tokens
- Access is controlled via role-based rules in each manifest
- Local directories manage agent registration in air-gapped setups
This design keeps sensitive operations local, minimizing risk.
What Are the Key Use Cases?
Some of the most promising industrial applications include:
- Predictive Maintenance – Agents detect faults and trigger MRO workflows
- Quality Feedback Loops – Vision agents alert MES systems when tolerances are off
- Energy Optimization – HVAC, lighting, and compressors coordinate to reduce load
- Autonomous Dispatching – Schedulers assign tasks to AGVs based on real-time data
- Simulation and Planning – Multi-agent models simulate production in digital twins
Is A2A Being Used in the Real World?
Early demos by Google focused on chatbot coordination. But adoption is expanding:
- Teams like Cursor use MCP to feed Slack conversations into agents
- Open standards groups such as CESMII, Eclipse BaSyx, and OpenOPC are exploring A2A as a coordination layer between industrial AI systems
These early steps suggest A2A is well-suited for on-prem innovation.
Final Thoughts
Protocols like OPC UA and MQTT solve the data transport problem. A2A adds a semantic layer for decision-making and collaboration.
Combined with MCP, it forms a flexible architecture for intelligent, tool-aware, and secure industrial automation.
For teams building autonomous, cloud-free environments, A2A provides a foundation for interoperable and evolving systems.
FAQ
Is A2A open source or proprietary?
A2A is an open protocol, initially released by Google, and can be implemented freely by industrial teams.
Can A2A run without an internet connection?
Yes. It is designed for air-gapped or edge-only environments and requires no cloud access.
What kind of agents can use A2A?
Any software agent with HTTP support can participate—vision models, SCADA connectors, analytics engines, and more.
