This is the foundation level of the Master AI Manufacturing Company Certification, rebuilt as a master class. It is written for the people who run and staff a small or mid-size plant: owners, plant and operations managers, quality managers and engineers, supervisors and shift leads, manufacturing, process and controls engineers, maintenance and reliability technicians, EHS coordinators, and the office staff who quote, purchase, plan and answer customers. It assumes you know your plant and nothing about artificial intelligence. A learner who finishes can decide, for one real task on one job, whether and how AI may be used, who owns the output, and how they would know if it stopped working.
The first two modules take apart what is sold to a plant as "AI". Module one separates fixed automation, rules, statistical process control, machine learning and generative AI, shows how a language model assembles an answer, and opens up the retrieval, tools and agents inside the products a small shop is pitched. Module two is about failure: standards, revisions and part numbers invented with complete fluency, tolerance and unit arithmetic that goes wrong silently, an accuracy figure read next to the recall it is hiding, and automation bias as the reason a habit beats a warning label.
Modules three and four cover the two data classes that make this trade different. Module three is export control — what counts as controlled technical data, why releasing it to a foreign person inside the United States can itself be an export, what the published-information and encryption provisions do and do not cover, and how to screen a file before it reaches any system. Module four is trade secrets and process knowledge: the statutory definition, the reasonable-measures element that a pasted prompt can quietly weaken, the notice employers owe, and a way to classify plant data before a tool sees it.
Modules five to eight are the duties that do not move. Module five names who owns each decision — the employer, the quality authority the system names, the licensed engineer where a state requires a seal, and a professional code that binds its members rather than the plant. Module six teaches law, law arriving through a contract, a voluntary consensus standard and a certification scheme apart, and places the quality standards and the voluntary AI management standard accurately. Module seven states plainly that OSHA has published no AI standard and no AI guidance, and teaches the general duty clause, machine guarding, hazardous energy, PPE, hazard communication and injury recordkeeping instead. Module eight separates a safety function from a perception feature, works through the robot-system failure modes OSHA itself documents, and explains why a vision system is never the safeguard.
Modules nine to eleven are about producing work you can check. Module nine builds a nine-part plant prompt around a supplied released document rather than a recalled one, starting with the data screen that happens before you type. Module ten builds a research workflow that confirms a standard, a regulation or an edition exists, is current, and reaches this product, this plant and this state, with a withdrawn quality standard edition as the worked example. Module eleven drafts work instructions, maintenance procedures, nonconformance narratives and supplier correspondence from a verified set, and marks where drafting stops: no tool drafts a disposition, a release or a conformance statement.
Modules twelve to fourteen build judgement and control. Module twelve reads detection claims like an engineer — accuracy, precision, recall, F1 and mean average precision, what the public datasets actually contain, a published miss rate converted into escaped defects on your own volume, and what a pilot on your own material has to look like. Module thirteen turns a one-off chat into a six-stage workflow with a trigger, a data-class screen, a named AI step, verification, a named approval and a record, and fixes where an agent's authority stops. Module fourteen draws the operational-technology boundary, sets the asset and access basics a plant without security staff can run, reads a vendor's data-handling claims for what they avoid saying, and shows how a federal contract clause becomes your security floor.
Module fifteen is a lab of four exercises on fictional plants where the errors are already in the material: an AI-drafted work instruction and nonconformance report with planted faults, a fabricated standard citation beside a wrong drawing revision, a quote package to screen for controlled technical data and trade secrets, and a vision-inspection claim to take apart. Module sixteen is the capstone, a Plant AI Implementation Plan for one low-risk first use, naming the data classes and controls, the duties that do not move, the vendor questions, and the baseline, failure indicators, stop conditions and improvement loop. It promises no scrap, downtime or productivity figure, because no source here supports one. A printable workbook and ten templates ship with the level.
Everything here is professional education. It is not engineering, quality, safety, cybersecurity, export-control or legal advice; it teaches no process parameter, tolerance, disposition or machine setting, and nothing in it runs on or near a control system. It does not replace a released document, a qualified engineer, the quality authority your system names, counsel, or the law of the state where the plant operates. Every statement of authority is labelled with whom it binds and when it was checked. Completing the level earns an independent educational certificate from AI Coalition Network with a public verification page. It is not an engineering licence, a safety or quality certification, a registrar's certificate or a regulatory approval, it carries no professional education hours, and it satisfies no licensing, registration or customer-qualification requirement.