From Industrial Software to AI Agents, AGENTIC: Sense, Reason, Act, Evolve

English

简体中文

English

简体中文

Appearance

From Industrial Software to AI AgentsFrom connecting devices to building bounded intelligent loops

This is not a conceptual book about bolting AI onto an IoT platform. It begins with the limits of industrial software and follows device access, data loops, cloud-native engineering, and safety governance to explain how agents can enter the physical world within deterministic boundaries.

Start Reading
Table of Contents
中文版
The question behind the book

Once a model can reason, how does it touch the real world safely?

In an industrial system, an answer cannot become an action directly. The field must first be observable, its signals must become trustworthy context, and every write must pass through an execution boundary.

01Observe

Make the physical world legible

Sensors, protocols, drivers, point values, and thing models turn fragmented field signals into stable platform data.

02Reason

Turn data into usable context

Data processing, rules, RAG, tools, and MCP give an agent the context it needs to understand a task and form a decision.

03Act

Put governance before execution

Identity, authorization, validation, policy, confirmation, recovery, and audit stand between model intent and a physical command.

An LLM can interpret intent and orchestrate governed tools. Real-time control, safety interlocks, and fault protection still belong to deterministic systems such as PLCs and edge controllers.

The narrative spine

The platform does not disappear when agents arrive. It becomes their foundation.

The book follows one cumulative engineering path rather than treating industrial software, IoT, and AI as three unrelated topics.

  1. 01Starting point

    Industrial software boundaries

    SCADA, DCS, MES, and PLCs solve essential local and deterministic problems, but data silos and closed capability boundaries limit wider collaboration.

  2. 02Platform base

    Unified IoT data and capabilities

    Multi-protocol access, thing models, cloud-edge collaboration, and platform data processing create a reusable digital base.

  3. 03Decision layer

    A governed Agent Runtime

    Spring AI, RAG, tools, MCP, state, workflow, and audit bring probabilistic reasoning into an explicit engineering boundary.

  4. 04Outcome

    Bounded autonomy

    The system can sense, reason, execute, verify, and recover without pretending that a model is a PLC or a safety system.

Reliable dataGoverned toolsDeterministic executionTraceable feedback
One concrete walkthrough

One cooling-pump vibration alert, end to end

The full mechanism chain in a single real incident: from a cooling-pump vibration value arriving to one confirmed write. Each step is developed in Chapters 4–5, 7, 8, and 14.

  1. 01

    Uplink & storage

    The driver reports the vibration point value over MQTT; unit and timestamp land in the time-series store.

    Ch. 4–5
  2. 02

    Threshold alert

    The rule engine lifts the excursion out of thousands of point values—deterministic logic holds the line first.

    Ch. 5
  3. 03

    Evidence, not guesswork

    Through authorized tools the agent reads trend, load, and maintenance records—the evidence points to early bearing wear.

    Ch. 7
  4. 04

    A pending Action

    Writes are never sent directly: a pending Action passes schema, permission, and risk checks first.

    Ch. 7–8
  5. 05

    Confirm & audit

    After human confirmation a deterministic path executes; receipts and audit trails stay queryable.

    Ch. 8, 14

Every step has a clear owner: data to drivers, alerts to rules, judgment to the model, actions to workflow, execution to deterministic paths.

Three parts · fourteen chapters

Build the complete loop, layer by layer

Each part inherits the constraints established by the one before it. The engineering base is never discarded when the story reaches AI.

IChapters 1–5

Foundations

From industrial software to the IoT platform base

Architecture, sensing, communication, and platform data processing: how physical signals become reliable data.

Enter Part I
IIChapters 6–9

Technology

Cloud-native and AI-agent capabilities

Microservices, containers, Agent Runtime, security, and standards: how intelligence enters the platform without bypassing its boundaries.

Enter Part II
IIIChapters 10–14

Applications

Multi-scenario practice and trusted collaboration

Industry, cities, agriculture, trusted data, and the IoT DC3 project: how the same base is tested against different constraints.

Enter Part III
What you should take away

Not a product manual—a reusable way to reason about systems

IoT DC3 is the open-source engineering reference that runs through parts of the book. The methods, boundaries, and trade-offs are the subject.

Architecture

Draw the full data loop

Explain the path from sensing and protocol access to storage, reasoning, command execution, and feedback.

Trade-offs

Choose with conditions attached

Compare protocols, cloud-edge boundaries, data paths, and security controls together with their costs and failure modes.

Practice

Design a bounded agent system

Define tool contracts, permissions, policies, confirmation, recovery, and audit before an agent can affect a device.

Three boundaries kept explicit

What this book does not claim

MCP replaces device protocols
MCP connects AI applications to tools and context; MQTT, CoAP, Modbus, and other protocols still connect devices.
A model directly controls equipment
High-risk writes require authorization, validation, policy, audit, and, where necessary, human confirmation.
IoT DC3 is the only answer
It is a concrete engineering reference used to make general architecture patterns and trade-offs inspectable.
Start with the real problem

Follow one signal from the field to a governed action

Begin with Chapter 1 to follow the full argument, or use the table of contents to enter at the layer closest to your current problem.

Start Chapter 1 Browse the contents
Not sure where to start? Read the guide first

From Industrial Software to AI Agents · Building a multi-protocol, cloud-native, open-source industrial IoT platform ready to evolve toward AI agents

By Zhang Hongyuan · © 2016–