Level 2 taught one person to do one plant task well with AI, under supervision, with a record. Level 3 is about the operation: the same work run by several people across parts, lines, shifts and customers, where the person accountable cannot read every output. It is written for plant and operations managers, quality managers, maintenance and reliability managers, continuous-improvement and industrial engineers, supply chain and materials managers, document control leads, and the IT or OT person who has to make it work on every shift.
The first four modules build the operating frame. Module one explains why a task that is safe once becomes risky on every shift and line: volume, handoffs, silent errors, workaround drift and complacency at the hundredth review, and why an operation needs a named owner who may pause it and a measured baseline before any redesign. Module two maps a plant workflow as it actually runs, including the undocumented steps and the person everyone asks, records every handoff, review point and record, marks each point where controlled technical data, controlled unclassified information or process secrets cross a boundary, and measures volume, cycle time, rework and escape. Module three tiers AI tasks by consequence and reversibility, attaches review depth, approver, record and escalation to each tier, keeps a tiering register with re-tier triggers, and raises the tier where the Colorado or Texas AI statute reaches a decision about a person. Module four writes an SOP a supervisor can enforce, in plain language the night shift will follow the same way, with approvers named by role, and tests it on someone who has never done the task.
Modules five to seven govern the knowledge and the access. Module five builds a retrieval corpus from controlled plant documents, screens export-controlled and controlled unclassified material out before ingestion because a corpus is a copy, tags every document by part, revision, customer, programme and effective date, and gives it an owner and a removal rule. Module six shows how a document assistant fails with confidence, through a withdrawn edition or superseded revision served as current, a near-neighbour part, another customer's requirements and instruction-like text inside a document, and pairs each failure with a citation rule and a check. Module seven keeps customer programmes in separate lanes of a shared tool, holds export-controlled material back from anyone not cleared, applies least privilege and strong sign-in, and logs access without creating a second uncontrolled copy.
Modules eight to eleven apply the frame to the shop floor. Module eight runs an AI-assisted inspection station as an operation, with a blind sampling plan that verifies the verifier, a monthly escaped-defect ledger built from measured recall, thresholds treated as controlled changes, and a record showing a named person owns every disposition. Module nine turns anomaly scores into a triage queue a named person works, deals with false alarms, missed failures and alarm fatigue, and keeps the hazardous-energy boundary that no alert crosses. Module ten gives free-text work-order history a checked structure, explains why that history is a poor training set, and drafts preventive maintenance procedures that state the energy-control and guarding boundaries. Module eleven is the honest module: no source in this programme measures AI in production scheduling, so it teaches method rather than performance, with the planner deciding, shortage analysis that ends at an engineering approval, and questions for a vendor claim with nothing published behind it.
Modules twelve to fourteen keep the operation honest over time. Module twelve builds a watchlist of the official sources that change plant work, each with an owner, reads new standard editions and scheme announcements for what they actually say, tracks each customer's requirements customer by customer, judges state and foreign AI laws by scope, and records what was checked, where and when. Module thirteen samples AI-assisted output instead of trusting it, sorts findings into an error taxonomy, sets thresholds against the plant's own baseline, feeds findings back into the SOP and prompt library, and measures the reviewer as well as the model. Module fourteen builds escalation paths that hold when a shipment is late, stop conditions agreed in advance, outside reporting clocks wired into the escalation matrix, and measures that reveal failure rather than reward review counts, and prepares the capstone.
Statements of authority are labelled with whom each binds and when it was checked; a voluntary consensus standard or certification scheme is never taught as law, and no state's rule is presented as a national one. The level ships with a printable workbook and ten templates, a final examination that draws forty scenario questions from a reviewed bank, and a capstone Workflow Redesign of one plant workflow built on fictional plants, parts, customers and figures.
Everything here is professional education. It is not legal, engineering, quality, safety, cybersecurity or export-control advice, and it does not replace a released document, the employer's safety programme, a licensed professional engineer where a state requires one, the person your quality system names, an empowered customer representative, your export authority, legal counsel, or the law that applies to your plant. Completing the level earns an independent educational certificate issued by AI Coalition Network with a public verification page. It is not an engineering licence, a quality-system registration, a safety qualification or a regulatory approval, it carries no professional development hours, and it qualifies no one as an auditor, a competent person or an authorised employee under any standard.