Lesson 17 of 20 · Chapter 17
The Technological Backbone: AI, 5G, and Blockchain
Up to now, software has told people what it saw. This chapter is about software that can act: an AI agent that plans the work, orders the parts and proposes the schedule. That amplifies everything, good and bad. The moment software can act, the plant must be even clearer about who decides, what the limits are, and what “done” means. The chapter adds two more pieces: a network fast enough to trust (5G), and a record nobody can quietly rewrite (blockchain).
Choose how you want to learn
You can change this any time. Every level does the same hands-on steps. Press Listen on any part to hear it read aloud.
- 1 · From inform to act
- 2 · The rule set
- 3 · 5G and speed
- 4 · The network is a basic condition
- 5 · A record you can trust
- 6 · Try it at work
- 7 · Your card
- 8 · Check yourself
Part 1
An agent coordinates; a tool only informs
An ordinary tool gives an answer. An AI agent works toward a goal across several steps.
Creating a work order when a number crosses a line is not agentic; it is just automating paperwork. An agent connects the evidence (camera, sensors, history), proposes a window that production can accept, checks the parts, and follows the job to proven closure.
It is the layer that joins everything up, so nobody has to do that in their head at 2:00 a.m.
A maintenance agent that cannot reach purchasing, stores and staging is like a vision system that detects defects but has no stop rules. If it identifies a likely seal head pad replacement, it should identify the parts, confirm inventory, and trigger procurement with urgency matched to risk.
It must also refuse to automate premature closure: if the evidence does not return to normal, it escalates rather than celebrates. And it respects “compensation is not restoration”: when the correction envelope is being used up, it recommends planned restoration, not more tuning.
“It is capable of moving from “inform” to “act,” and the moment software can act, the plant must be even clearer about decision rights, guardrails, and what “done” means.”
“That alone is not agentic; it is automation of a clerical step.”
“Agentic AI is the layer that can connect these streams without forcing people to do the mental integration at 2:00 a.m.”
Read the whole chapter free in the book reader.
Six things a maintenance system does. Which are plain automation, and which are agentic?
Part 2
What the agent may do, and what it may never do
A safe set of rules has four parts.
1. It may watch and recommend freely, as long as it logs its reasoning. 2. It may draft work orders and propose times, but a person approves. 3. It may do low-risk admin on its own: reserve parts, notify people.
4. It may never, on its own, change how the process behaves: setpoints, stop rules, defect thresholds, or safety interlocks.
The agent is part of the plant’s controlled release of normal, the same idea that warned against silent threshold changes. Its proposals must stay inside approved windows, respect safety boundaries, and be honest about uncertainty.
It must explain itself in the plant’s language, not “the model predicts failure” but evidence people can check: correction frequency up threefold in 48 hours, current variance above normal, a similar pattern before the last rebuild.
“First, the agent may observe and recommend freely, as long as it logs its reasoning and the data behind it.”
“Second, the agent may create work objects and propose schedules, but a human owns approval.”
“Fourth, any change that affects process behavior, such as modifying setpoints, changing stop rules, widening defect thresholds, or overriding safety interlocks, is never autonomous by default.”
Seven actions an agent could take. Where does each one sit in the rule set?
Part 3
Control loops close locally; learning loops close globally
An agent can plan work, but it cannot bend time. Some decisions must happen in a fraction of a second.
If a camera’s images travel to a distant server over a busy network, the delay grows and varies. The line reacts too late (scrap) or too often (needless stops).
So the rule is: the fast control loop (detect, correct, check) runs right at the machine. The slower learning loop (trends, comparisons, retraining) can run anywhere.
A private 5G network can give reliable, time-synchronised connection across a wide sensor footprint. That reliability turns “maybe the sensor glitched” into “the signal is trustworthy,” which is what production needs before honouring a planned window.
5G also supports mobility (people moving between stations with consistent access) and dense connection, so making abnormal visible becomes the default state rather than a special project.
“Agentic AI can coordinate maintenance work, but it cannot bend time.”
“If the image stream or defect classification has to travel to a distant server over a congested network, latency grows and variability grows.”
“This leads to a design principle that fits modern TPM: control loops close locally; learning loops close globally.”
Six jobs the system does. Which must close locally, at the machine, and which can close globally?
Part 4
A flaky network is normalised abnormality
If the network is needed to check and close work, then the network is a basic condition, just like a clean lens or steady lighting.
When a screen lags or a device drops out, nobody should shrug and say “the network is flaky.” That is an abnormality becoming normal.
Network health needs owners, simple indicators, and a response: find it, fix it, prove it, learn from it.
Treating 5G as “IT’s network” rather than part of the operating system is where it fails. Security and trust must be designed in: who can see what, what is logged, what is aggregated, what is used for immediate protection versus long-term redesign.
A real-time plant must be a trusted plant, or the workforce responds with silence and workarounds, just faster.
“When a dashboard lags or a device drops, the plant should not shrug and accept “the network is flaky” as normal. That is normalized abnormality at the infrastructure layer.”
“First, network reliability becomes a basic condition, the same way clean optics and stable lighting are basic conditions for vision in Chapter 14.1.”
“A real-time plant must be a trusted plant, or the workforce will respond with silence and workarounds, just faster.”
Three network moments on the plant floor. What would the chapter do?
Part 5
The record is part of the control system
Faster action raises a new risk: can you trust the record? A check “ticked off” later to improve compliance numbers poisons everything built on it.
Blockchain here is not a buzzword. It makes the maintenance record tamper-evident: corrections become new entries, and the evidence at closure is fingerprinted so nobody can quietly change it later.
You do not put everything “on the blockchain.” You store fingerprints and references. The bench below puts all three pieces together.
The problem blockchain solves is not storage but chain of custody: agent proposals, parts reservations, interventions and verification, linked and unalterable. Its strongest uses are where several parties must trust one record: sites comparing failure modes, auditors checking calibration, spare-parts genealogy against counterfeit or wrong-spec parts.
The same rule binds humans and agents: if the verification gate is not complete, the record cannot declare victory. And governance still matters: a tamper-evident record used as a weapon against individuals destroys psychological safety.
“The practical problem blockchain solves in maintenance is not storage; it is chain of custody.”
“A tamper-evident record does not make people virtuous. It makes it harder for the system to drift into self-deception.”
“In modern TPM, the maintenance record is not a diary. It is part of the control system.”
“If the record is trustworthy, the plant can learn once. If it is not, the plant will relearn the same lesson forever, only faster.”
The Agent Rulebook. An AI agent watches the sealing station. Set what it may do on its own, and how it checks, explains, closes, connects and records.
Part 6
Try it where you work
Do these on a real machine: at work if you work with equipment, or at home on a car, washing machine, furnace or lawn mower if you do not. Tick each one when it is done. Safety first: look and clean only what you are allowed to touch, never reach into moving equipment, and follow your site’s lockout rules.
Part 7
Your “My Agent Rules” card
The rules for one AI agent before it is allowed to act. Nothing you type here leaves this device.
My Agent Rules
Global Sovereign University · Modern TPM, lesson 17
The backbone
- Agents coordinate; people decide.
- Observe freely; act only on low-risk admin.
- Never change process behaviour on its own.
- Control loops local; learning loops global.
- The network is a basic condition.
- Append-only records; corrections are new events.
The moment software can act, be clearer about who decides. Learn once.
Part 8
Check yourself
Six real situations. Pick what you would do. You will see why after each one.
Lesson 17 complete.
You can now tell automation from an agent, write the four-part rule set, decide which loops must close at the machine, treat the network as a basic condition, and keep a record nobody can quietly rewrite. Next: Lesson 18, measuring TPM success: OEE and KPIs.
Modern Total Productive Maintenance (TPM) by Dr. Gene A Constant. This lesson comes from chapter seventeen. Read the whole book free in the reader, or keep a copy.
Buy a copy — it funds a free seat
GSU is a 501(c)(3) nonprofit and book sales are its only income. A purchase is not a tax-deductible donation. This lesson is free and always will be.