Lesson objective
Plan a plant AI session inside the release window a supervisor can actually give, with a minute-by-minute budget that keeps practice ahead of presentation.
Plain-language explanation
Why this matters now
Every plant has a release window, and it is usually smaller than the instructor wants. A supervisor can let four people go for twenty minutes before the shift, or the whole cell for thirty minutes during a scheduled changeover, or everyone for an hour on a planned down day twice a year. An instructor who designs a ninety-minute session and then asks for the time designs a session that never runs, or runs once for the people who could be spared, who are rarely the people who most need it.
This lesson is about working from the window, not towards it. It rests on no research about ideal session length, because this programme holds none. The lengths and budgets below are this programme's recommendations, based on how plants release people, and you should adjust them to how your plant actually runs.
A question to sit with
Ask three supervisors how long they could release their people for an AI session without hurting the shift. If their answers are fifteen, twenty and thirty minutes, what does your ninety-minute plan become?
What goes wrong
1. The session is designed first and the time requested second. The plan needs ninety minutes; the plant offers twenty; the instructor cuts from the end, which is usually where the practice was.
2. Presentation eats the practice. In a short slot, the introduction, the context and the tool tour expand to fill the time, and the checking activity is "skipped for time".
3. One session, one shift. First shift gets the session. Second and third shifts get the slides by email. The behaviour changes, if at all, on one shift out of three.
4. The AI session is folded into required training to save time. A supervisor suggests running it inside the annual lockout refresher "since everyone's already there". The required training is diluted and the AI session is filed as something it is not.
5. The session is timed for a calm room. The plan assumes nobody is called out. In a plant, somebody always is.
The practical answer
Best practice: find the real window first. Before designing, ask the supervisors who will release people three questions: how long, when in the shift, and how many at a time. Write the answers down. The window is a design input, like the audience.
Best practice: fit one behaviour to one short segment. Where the window is fifteen to thirty minutes, this programme recommends one behaviour per segment, not a compressed version of the whole session. Four twenty-minute segments across a week, each on one behaviour, reach more people than one session nobody can attend.
A worked twenty-minute budget. One behaviour, one fictional document, one check.
| Minutes | Element | Notes |
|---|---|---|
| 2 | Opening: what this session is and is not | One sentence, including that it is not required training |
| 3 | The fictional AI output on screen and on paper | "Would you let this go out?" |
| 8 | The check, done by every learner | Against the fictional source record, with the planted error |
| 4 | Debrief tied to the behaviour | "What would have caught it on your shift?" |
| 3 | The three things that never get cut | Data screen, named approver, verification step (Lesson 4) |
| 20 | Total | Practice (8 minutes) is the largest single block |
The rule of thumb this programme recommends: the checking activity is the largest single block in any segment, and nothing is added ahead of it once the plan is set. If the room runs late, cut the debrief short, not the check.
Why practice under some time pressure belongs in the plan
Research (Goddard, Roudsari and Wyatt, Journal of the American Medical Informatics Association, 2012): this systematic review of clinical decision support in healthcare, before large language models, names "workload, task complexity, and time constraint" among the environmental factors affecting automation bias. In a plant, the check will be done under a shift's time pressure, not in a calm training room. A timed repeat of the check, where the window allows, is this programme's way of rehearsing that. The review studied healthcare; the plant application is an analogy, and it says nothing about ideal session length.
Professional guidance (voluntary): NIST AI 600-1 (July 2024) lists "over-reliance" and "automation bias" within its Human-AI Configuration risk. It gives no training design; it is a reason, not a method.
Best practice: run the same segment on every shift. Plan from the start to deliver the segment on first, second and third shift, with the same fictional materials and the same planted errors. Module 4 covers facilitating the same session on every shift. At the design stage, what matters is that the plan assumes it.
Best practice: plan for the call-out. Somebody will be called back to the floor. Design each segment so that a learner who leaves after the check has still done the part that matters. Put the check before the debrief, not after a long introduction.
Best practice: book the room off the floor. A release window of twenty minutes is not a reason to teach at the cell. Hold the segment in a training room, break room or office away from running equipment. Module 7 covers teaching safely around production.
Best practice: record what was delivered to whom. For each segment, note the date, the shift, the attendees, the behaviour taught, the materials used and the sources cited. That record is what lets a correction reach the right people later, which Module 3 covers.
Remember this
- Find the release window before designing; it is a design input, not an afterthought.
- In a short window, teach one behaviour per segment; the checking activity is the largest single block.
- Never fold an AI session into required training such as the lockout programme's employee training, and never take its time.
- Run the same segment on every shift, off the floor, and design for the call-out.
- Session lengths here are this programme's recommendations, not findings.
Manufacturing example
Sorrell Metal Finishing, a fictional anodising and plating shop running three shifts, wants its operators trained on checking AI-drafted process-sheet summaries. The fictional training coordinator, Jonah Ekwueme, drafts a seventy-five-minute session. He then asks the three shift supervisors for their windows. First shift can release six people for twenty minutes before start-up. Second shift offers thirty minutes during a scheduled tank changeover, but only four people at a time. Third shift says twenty minutes, twice a week. One supervisor adds: "Why not run it inside the lockout refresher next month? Everyone's there anyway." Jonah declines that at once, explaining that the lockout refresher is part of the employer's energy-control programme, owned by the fictional EHS lead, and the AI session must neither dilute it nor be filed as part of it. He redesigns the content as three twenty-minute segments, one behaviour each. The first segment's budget is two minutes of opening, three minutes on a fictional AI summary of a fictional process sheet, eight minutes for each operator to check it against the fictional source sheet, where the summary has swapped two bath temperatures, four minutes of debrief and three minutes on the data screen, the named approver and the verification step. He runs it in the break room, never at the line. On second shift, two operators are called out at minute twelve; both had already done the check. Jonah records each segment's date, shift, attendees, materials and sources in his session log, so that any later correction can reach the same people.
Responsible-use warning
Never design a long session and then ask for the time, and never cut the checking activity to save minutes. Never fold an AI session into the lockout programme's employee training or any other required training, take its time, or record it as part of it. Never teach at the cell because the window is short. And never justify a session length, a segment count or a sequence with a claimed research finding; the budgets here are this programme's recommendations.
Human-review requirement
The plant manager decides how much production time an AI session may take, and each shift supervisor confirms the release window. The employer, through the EHS lead or whoever its energy-control programme names, owns lockout training and confirms the AI session takes none of its time or content. The instructor designs inside the window, records what was delivered to whom, and reports in writing when a window is too small for a behaviour that matters.
Practical exercise
Fit a session into real release windows. Use a fictional plant and fictional shifts throughout.
- Describe a fictional plant with at least two shifts and write down the release window each supervisor can offer: how long, when in the shift and how many people at a time.
- Take three behaviour statements from Lesson 1 and decide how many segments you need to cover them inside those windows.
- For one segment, write a minute-by-minute budget that adds up exactly to the window, with the checking activity as the largest single block.
- Add up your budget again and confirm the total matches the window.
- Write what you will cut first if the room starts late, and confirm it is not the check.
- Describe where the segment will be held, away from running equipment, and how the same segment will run on every shift.
- A supervisor proposes folding the session into the lockout refresher; write your reply in two or three sentences, naming who owns that training.
- Design the session log entry you will keep for each delivery: date, shift, attendees, behaviour, materials and sources.
- Write the short note you would send the plant manager if one behaviour needs more time than any supervisor can release, stating what is at risk and what you recommend.
Knowledge check
-
Why should an instructor find the plant's release window before designing a session?
Answer guidance: Because a session designed first and fitted later loses its practice to cuts, or never runs; the window is a design input, like the audience, and sets how many segments the content needs.
-
What does 29 CFR 1910.147(a)(3) require, and why does it matter to AI session planning?
Answer guidance: It requires employers to establish an energy-control programme of procedures and employee training; that training is its own programme, so an AI session must never be folded into it, take its time or be recorded as part of it.
-
In this programme's twenty-minute budget, which element is the largest single block and why?
Answer guidance: The check done by every learner, eight of the twenty minutes, because practising the behaviour is what the segment exists for; if the room runs late, the debrief is cut, not the check.
-
Is a twenty-minute segment length a research finding?
Answer guidance: No. This programme holds no research on ideal session length; segment lengths and budgets are its recommendations, based on how plants release people, and should be adjusted to the plant.
Sources and further reading
- Occupational Safety and Health Administration (U.S. Department of Labor) — 29 CFR 1910.147 - The control of hazardous energy (lockout/tagout). Accessed 2026-09-15.
- Kate Goddard, Abdul Roudsari and Jeremy C. Wyatt; Journal of the American Medical Informatics Association — Automation bias: a systematic review of frequency, effect mediators, and mitigators (2012-01-01). Accessed 2026-09-15.
- National Institute of Standards and Technology — Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile, NIST AI 600-1 (2024-07-26). Accessed 2026-09-15.
- AI Coalition Network — Certification Educational Standard: Human Review of AI-Assisted Professional Work. Accessed 2026-09-15.