Andragogy vs Pedagogy: Differences and When to Use Each

Go from storyboard to an interactive course in minutes

Start Free

A new safety procedure lands on a training team's desk with one line of guidance: "They're all adults, so keep it andragogical." Half the people who need it have done this work for years and will switch off at a step-by-step walk-through. The other half have never seen the equipment, and within a month they will be acting on the procedure alone. A course built on one set of assumptions about its learners will let one of those groups down.

I'm an instructional designer at Coursebox, where we make instructional design software that turns a brief into a course outline, so I'll mention our product where I tested it. Malcolm Knowles used andragogy and pedagogy to name two sets of assumptions about learners, and the useful question is which set is realistic for this group on this content, not how old the learners are. Here I set out what each term means, what Knowles actually compared, where heutagogy fits, two real studies, a test of one procedure briefed for two groups in our product, and two free templates.

What is andragogy?

Andragogy is Malcolm Knowles' model of assumptions about adult learners: that they move towards self-direction, bring experience that is a resource for learning, become ready to learn when a real-life task needs it and focus on performance rather than subjects.

View the full-size diagram

The word comes from the Greek stem andr-, which Knowles glossed as "man, not boy" or adult, and the second half means leading. Knowles did not coin it, but he is the reason it is widely used in English. In the 1980 revision of The Modern Practice of Adult Education he looked back on his first definition: "Originally I defined andragogy as the art and science of helping adults learn, in contrast to pedagogy as the art and science of teaching children."

Then he changed his mind. Schoolteachers had been applying his ideas to young people and reported that "in certain situations they were producing superior learning". So by 1980 he saw andragogy as "another model of assumptions about learners to be used alongside the pedagogical model of assumptions". The two models are ways of testing which assumptions "fit" a particular group in a particular situation, not labels for two age groups.

For this comparison, it helps to separate three things:

• The assumptions. Four in 1980, six in current editions. These are what the comparison with pedagogy is about.
• The design process. Knowles' separate seven-phase process for planning learning with adults, and his learning contract. Our guide to adult learning theory covers both, with the six assumptions in depth.
• The comparison. Pedagogy and andragogy as two sets of assumptions to choose between for each group and each piece of content.

Here are the two terms at a glance, with heutagogy, a third term from 2000. The table is my summary of the sources quoted below, not Knowles' own table.

What does pedagogy mean?

Pedagogy means the art, science or profession of teaching, for learners of any age; the narrower sense of "teaching children" comes from the word's Greek roots, which Knowles used as a contrast when he introduced andragogy.

Merriam-Webster defines pedagogy as "the art, science, or profession of teaching", and traces it to the Greek paidagōgos, "originally the word for an enslaved person who brought children to school". The paidagōgos was a household attendant who walked a boy to his lessons, not the teacher.

Knowles read the roots literally. Pedagogy, he wrote, is "a term derived from the Greek words paid (meaning "child") and agogus (meaning "leading"). So "pedagogy" means, literally, the art and science of teaching children." He traced the pedagogical model to the monastic schools of Europe between the seventh and twelfth centuries, where its assumptions were "based initially on observations by the monks in teaching very young children relatively simple skills". That history is Knowles' account, and it explains his contrast. It does not change what the word means in use.

In everyday use, teaching is what happens in the room and pedagogy is the approach and reasoning behind it. Outside English the word is broader still. Loeng's 2013 study of Eugen Rosenstock-Huessy describes him leading the first academic thinking on "adult pedagogical issues in the 1920s", a phrase that would make no sense if pedagogy only meant teaching children.

You will also see lists of "five types of pedagogy", such as constructivist, collaborative, inquiry-based, reflective and integrative, or four types built on behaviourism, constructivism, social constructivism and liberationism. I found no authority behind any fixed count. These are groupings of teaching approaches, and each list is someone's selection.

Where the word andragogy came from

Knowles did not invent the word. Its first known use is by a German schoolteacher, Alexander Kapp, in 1833, in a book on Plato's educational ideas. As Loeng's 2017 study of Kapp puts it, "This is not a theory of adult learning." Kapp used it to name a part of his book about education in adulthood, and the word was then largely forgotten.

Eugen Rosenstock revived the word in the 1920s while teaching at the Academy of Labour in Frankfurt. Eduard Lindeman's 1926 book The Meaning of Adult Education never uses the word andragogy, though it states much of what Knowles later built on, such as "Experience is the adult learner's living textbook." Lindeman did use the word in 1927, in a book written with Martha Anderson, but he did not develop it.

Knowles learned the label in the mid-1960s from a Yugoslav adult educator, Dušan Savićević, and used it in Adult Leadership in April 1968, printed with the misspelling "Androgogy". His book The Modern Practice of Adult Education followed in 1970 with the subtitle Andragogy versus Pedagogy. The 1980 revision changed it to From Pedagogy to Andragogy, and that change of subtitle tracks the change in his thinking.

So who is the father of andragogy? Knowles is usually given the title because he popularised the word in English and built a model around it. Kapp used it first, and Rosenstock and Savićević carried it in Europe before Knowles took it up.

Which books to read

• Knowles, The Modern Practice of Adult Education: From Pedagogy to Andragogy (1980 revision). Read chapter 4. In about 20 pages it gives his comparison table, the spectrum passage and both of his age examples.
• Knowles, Holton and Swanson, The Adult Learner. Read it for the six assumptions and the research since Knowles. The ninth edition (Routledge, 2020) adds Robinson as a fourth author.
• Hase and Kenyon, "From andragogy to heutagogy" (2000). Read it for the third term. It is short and free through Southern Cross University's research portal.

Pedagogy vs andragogy: what Knowles actually compared

The best primary source for the difference between pedagogy and andragogy is Exhibit 4 in the 1980 book, titled "A comparison of the assumptions of pedagogy and andragogy". It has four rows, and each row sets a pedagogical assumption beside an andragogical one. The two later assumptions are not in it.

Read the pedagogy column closely. "Most people of the same age are ready to learn the same things" describes an age-graded school, with "a fairly standardized curriculum, with a uniform step-by-step progression for all learners". That is one model of teaching, not the whole of teaching. Knowles listed the techniques he associated with each side: "transmittal techniques" for pedagogy, and for andragogy "experiential techniques" such as "discussion, problem-solving cases, simulation exercises, field experience".

Both groups in the example below are ready for a real task and performance-focused. Their routes mainly differ in prior knowledge and the support needed. The contrast is also not between good and bad teaching. Asked whether research showed andragogy was superior, Knowles began his answer with "That is not the question".

Two ends of a spectrum, not children and adults

Knowles' co-authors later wrote that the 1970 book presented the models as if "pedagogy is for children and andragogy is for adults". By 1980 Knowles saw them as most useful "not as dichotomous but rather as two ends of a spectrum", with the realistic assumption for a given situation somewhere in between.

He gave two examples, one about a child and one about an adult: "a six-year-old may be highly self-directing in learning the rules of a game but quite dependent in learning to use a calculator; on the other hand, a forty-year-old may be very dependent in learning to program a computer but completely self-directing in learning to repair a piece of furniture."

From that he drew a rule for designers: "whenever a pedagogical assumption is the realistic one, then pedagogical strategies are appropriate, regardless of the age of the learner". The rule runs the other way too: where an andragogical assumption is realistic, andragogical strategies fit, whatever the learner's age. In 1979 he applied the test to himself. He had meant the assumptions "to be tested (not to be presumed)", and "if I were now, at age 66, to undertake to learn a body of totally strange content (for example, the higher mathematics of nuclear physics), I would be a totally dependent learner."

He added a warning that is easy to miss. A designer committed to pedagogy as an ideology "may be tempted to underrate the extent to which an andragogical assumption may be realistic and may, for example, want to keep a learner dependent long after the learner has become able to be self-directing." So the spectrum cuts both ways: start where the learner is on this content, and plan to move them along as they become able to steer.

A standard textbook on adult learning by Sharan Merriam and colleagues reaches the same view: andragogy "now appears to be situation-specific and not unique to adults". In practice, the question for every course is where these learners sit on this content, not whether they are adults.

Even a course designed on Knowles' model bends once it meets a workplace, as a Canadian study shows.

Case study: Communication training for paratransit drivers in Quebec

This is a designer's own account of applying Knowles' andragogical process model at work, with a small before-and-after study of the results. An adapted transport service in a large Quebec city employed 145 unionised drivers and made more than three million trips a year, but its driver training included "no specific training" on communicating with passengers who have communication disabilities, such as aphasia.

• What they did: In 2016 and 2017 a speech-language pathology researcher designed a 5.5-hour training in two sessions, 2 to 5 days apart so drivers could try the strategies on the road, using the eight-element version of Knowles' process model from a 2015 edition of The Adult Learner. Her needs analysis drew on six drivers, the chief operating officer, four passengers and three and a half days of observation, including riding along. Thirteen volunteer drivers took part, aged 33 to 57, with between 4 months and 11 years on the job.
• Evidence and result: In Tessier, Croteau and Voyer's 2021 paper in the Journal of Workplace Learning, only three of the model's eight elements were fully applied. The trainer set the objectives because she "was the best person to elaborate training objectives", and they "were neither discussed nor negotiated with each participant". Needs were not assessed person by person because drivers were hard to release from work. Only 53.8% of the 13 drivers strongly agreed that the training matched their needs, and fit with needs was among the lowest-rated items. A companion paper in Aphasiology reports that scores on a written scenario test rose after training, and that videos of real trips showed significant changes in drivers' communication.
• Limitation: Thirteen volunteers at one service, with no control group. The trainer designed, delivered and evaluated the training and knew the drivers, the satisfaction ratings are self-reports, and I read the outcome paper as an abstract only.
• Lesson: In this service, the elements that depend on each learner, negotiated objectives and individual needs diagnosis, were the first to go, and fit with needs was among the items drivers rated lowest. If you cannot negotiate objectives with each learner, at least test them against a few learners' real situations before you build.

The six assumptions of andragogy, in brief

Knowles' list grew over time, which is why sources disagree on the number. His co-authors on The Adult Learner put it plainly: "the number of assumptions has grown from 4 to 6 over the years".

So a list of five is the version from 1984 until the need to know was added in 1989 and 1990, and a "1968" or "1970" list of six is wrong. Here are the six in the order of current editions, with the row of the 1980 comparison each one comes from.

For the comparison, the four 1980 rows are enough.

When pedagogy is the right choice for adults

Knowles' rule makes pedagogical strategies a legitimate choice for adults whenever the pedagogical assumptions are the realistic ones. For a workplace course, these are the signs I look for:

• The learners are new to this content. Not to work in general, but to this task, system or procedure.
• Errors are costly or unsafe. A wrong first action could hurt someone, damage equipment or breach a rule.
• They will act alone soon. There is no experienced colleague beside them on the day.
• The task gives little feedback. Learners cannot tell from the task itself whether they did it right.
• There is a required performance standard. Everyone must meet the procedure. That constrains the outcome, but learners can still help plan how they reach it.

The research points the same way, with limits. Cognitive load researchers call one pattern the expertise reversal effect: a technique that helps learners with little prior knowledge, such as a fully worked example, can stop helping or even hinder learners who already know the content. Kalyuga's 2007 review of the expertise reversal effect traces the research, though I read it as an abstract only, and our guide to cognitive load theory explains how to plan for it. In a study of people learning to search for information electronically, Debowski, Wood and Bandura found guided exploration beat self-guided exploration, and put it down to uninformative feedback, which "causes most people to remain at a novice level and to require external guidance". I read that study as an abstract only.

Giving adults control is also a design bet, not a given. In Brown's 2001 study of 78 employees taking an intranet course with learner control, the amount of practice and time on task varied widely between trainees, and both predicted how much they learned. In Fisher and colleagues' 2017 study of 518 trainees in a Fortune 500 organisation, the relationship between the control learners said they wanted and what they actually did with it "was relatively weak". I read both as abstracts only, so I could not see what the training covered.

Pedagogical does not have to mean lecture first. In a 2024 randomised trial with emergency medicine residents, working through hard head CT cases that faculty had chosen, before any instruction, improved scores on unfamiliar cases, most of all for residents earlier in training, with no statistically significant difference on routine ones. I read Knowles' pedagogical assumption as being about who decides and who carries the structure, not about whether explanation or practice comes first.

I found no workplace study that compares an andragogical and a pedagogical version of the same course for novices on safety-critical content, with results measured on the job. So the checklist rests on Knowles' own test and these studies, not on a direct trial.

Andragogy vs pedagogy in practice: one procedure, two groups

Here is the hypothetical example I'll use for the rest of this guide. A private operator runs six indoor aquatic and leisure centres. From 14 April, every centre moves to a new automatic chemical dosing controller with a new alarm screen and a phone alert. Two groups must learn the same procedure, "Responding to a dosing alarm on the new controller".

• Group A, 8 senior plant operators. Each has run pool plant rooms for 5 or more years, knows pool water chemistry and the old controller, and is on the after-hours call roster. What is new for them is the screen layout, the alarm codes, the controller's automatic dosing lockout and the phone alert.
• Group B, 15 newly appointed duty supervisors. Most are former lifeguards, and none has worked in a plant room. From 14 April they are the first responders to a dosing alarm on evening and weekend shifts when no plant operator is on site. Their job is to make the pool and people safe and call the on-call plant operator, never to adjust dosing or handle chemicals.

Every fact the courses use is the operator's own site procedure, not a regulation or chemistry advice. The controller shows five alarm codes: CL-LOW and CL-HIGH (free chlorine outside the site range shown on the screen), PH-LOW and PH-HIGH (pH outside the site range), FLOW (no water through the sample cell, so the controller stops dosing automatically) and GAS (the plant room chlorine gas detector).

• GAS: nobody enters the plant room; clear the pool hall and plant area using the site emergency plan; call emergency services and the on-call plant operator.
• CL-HIGH: close the pool to swimmers and call the on-call plant operator.
• Any other alarm: duty supervisors call the on-call plant operator and keep the pool open only if the operator says so; plant operators diagnose and correct.
• Always: duty supervisors never adjust dosing settings or handle chemical containers, nobody mixes chemicals, and every alarm goes in the plant log with the time, alarm code, reading on screen, action taken, and who was called and when.

The courses use the alarm screen, a plant log page, the on-call roster card, an extract from the site emergency plan and three scenarios: a Tuesday-morning CL-LOW during lap swimming, a Saturday-night CL-HIGH during a children's party and a GAS alarm at 6:15 am before opening. By 14 April, plant operators should be able to respond to and correct any of the five alarms and coach a duty supervisor by phone. Duty supervisors should be able to recognise each alarm, take the right first action within their limits, call the operator with the right information and complete the log entry. A plant operator then signs each duty supervisor off on a supervised shift before their first solo shift. The course does not make anyone a plant operator.

This example shows Knowles' point because the content and the date are identical, while the two groups sit at different positions on some of his assumptions for reasons that have nothing to do with age. A 45-year-old new duty supervisor is a novice here, and a 26-year-old plant operator is not. Many of the duty supervisors are self-directing adults in other parts of the job, such as rescues, first aid and complaints, which is Knowles' forty-year-old example made concrete.

1. Concept of the learner

The first row asks how much the learners can steer on this content: dependent on the course for what comes next, or able to decide for themselves.

• Plant operators: Self-directing on this content. Their prior work is relevant, but an entry scenario should confirm what they know about the new controller before skim advice is offered.
• Duty supervisors: Dependent on this content. They cannot yet tell which parts of an alarm screen determine their first action, so choice would leave them guessing.
• Design decision: Operators get a self-check that tells them what they can skim. Supervisors get a fixed order with no skipping of the GAS and CL-HIGH content.

2. Role of learners' experience

The second row asks whether the learners' own experience is something the course can build on, for this content.

• Plant operators: Rich and relevant. In this hypothetical group, every operator has handled alarms on the old controller, and some habits will need replacing on the new one.
• Duty supervisors: Little plant experience, but real emergency experience from lifeguarding, such as clearing a pool hall.
• Design decision: Operators get an open-text activity on an alarm they handled on the old controller and what they would do differently now. Supervisors' activities draw on clearing the pool hall, not on plant knowledge they do not have.

3. Readiness to learn

The third row asks whether the learners already feel the need for this, or whether the course has to create it.

• Plant operators: Ready because of 14 April. The new controller changes their own work, and they will want the differences, not a full course.
• Duty supervisors: Ready in principle, because they will be first responders, but they cannot yet picture what an alarm at night will ask of them.
• Design decision: Operators start from what changes on 14 April and what stays the same. Supervisors start with their limits and a picture of the first alarm they will face alone.

4. Orientation to learning

The fourth row asks whether the course should be built around tasks or around topics, and in what order.

• Plant operators: Performance-centred. They want the alarm first and the screen explanation only where they need it.
• Duty supervisors: Also performance-centred, but they need to see the task done before they try it.
• Design decision: Operators open on the Tuesday CL-LOW scenario before any explanation. Supervisors see a fully worked CL-HIGH response, showing the screen, the first action, the call and the log entry, before any question.

How to choose between andragogy and pedagogy for a course

Knowles left a test, not a method, so this workflow is my own, built on his comparison and carried through the dosing alarm course.

1. List the groups who need the content. Here, plant operators and duty supervisors.
2. For each group, answer the four 1980 rows for this content. Can they steer? Is their experience a resource? Do they already feel the need? Is their orientation performance-focused or subject-focused?
3. Check the risk. Are errors costly or unsafe? Will the learners act alone soon? Does the task give them little feedback on whether they got it right? If the answer is yes, lean pedagogical for anyone new to the content, whatever their age.
4. Choose a route per group. For these operators: task first, checked skim advice and their own experience. For these supervisors: a walk-through, worked examples and a fixed practice order. Both routes need checks that fit their responsibilities and risk; these formats are design choices, not definitions of the models.
5. Plan the move along the spectrum. Say what will show that a learner can steer, and what changes for them when they can. Knowles warned against keeping learners dependent after they no longer need it.
6. Write it into the brief. Name each group's position on the content and the route that follows, not only their job title.
7. Check the work, not only the course. For the duty supervisors, that is the supervised shift. A training needs analysis is the larger version of step 2, and the behaviour level of Kirkpatrick's evaluation model is the larger version of step 7.

The two-route plan for the dosing alarm course

Here is the plan for the two groups. The timings and the sign-off are this operator's choices, not rules from Knowles.

The worked example on the supervisors' route does the job that Merrill's principles of instruction call demonstration, and the scenarios on both routes follow the pattern of scenario-based learning.

Moving a learner from one route to the other

Knowles warned against keeping a learner dependent after they can steer, so expect a learner's position on the content to change. A duty supervisor who has handled several alarms with the on-call operator may be ready for more choice on CL-LOW, if a fresh scenario and supervised work confirm it, and a refresher can let them skip what they have shown they can do. A plant operator facing a new FLOW lockout they have never seen may want the walk-through for that one part. Write down what evidence moves a learner along, such as a passed scenario or a signed-off shift, and who decides.

What course briefs say about learners

To see how people describe learners when they brief a course, I looked at course requests typed into Coursebox in the 90 days to 7 October 2026: one per person, from English-language browsers, with no uploaded file. Of 266 prompts, 104 were course requests, and 24 to 27 of those were for workplace courses. Two independent AI readers each classified all 104 against written definitions, every positive call was re-read by hand against the request, and borderline calls count only in the upper figures.

The only requests classified as adapting a course to the learner's level were for school pupils and one homeschooled child. The classification identified 1 clear request and 1 borderline request using the word adult for its learners, and the one workplace request that gave an age range gave it as a description of the audience, not as a reason to change the design. About 1 in 5 described learners by age or school level, and 11 of those were school pupils.

So briefs say who the learners are, by role, group or age, but almost never where they sit on Knowles' spectrum for this content. The sample is small, many prompts are cut off at 1,500 characters in the export, and it covers requests, not finished courses. Organisations may also run separate courses for new and experienced staff that never appear as one brief. Still, a course builder, human or AI, can only work from the brief, so if two groups need different routes, the brief has to say so.

Heutagogy: the third term

Heutagogy is the study of self-determined learning, a term Hase and Kenyon introduced in December 2000 for learning where the learner, not the teacher, decides what to learn, how, and how to judge it.

Hase and Kenyon set it out in "From andragogy to heutagogy", in the RMIT journal ultiBASE. They wrote that andragogy "still has connotations of a teacher-learner relationship", and that heutagogy is not "a departure from andragogy but, rather an extension that incorporates self-directed learning". In their picture, "the teacher provides resources but the learner designs the actual course he or she might take by negotiating the learning", and assessment "becomes more of a learning experience rather than a means to measure attainment". It was a position paper, with no data.

Blaschke's 2012 review of heutagogical practice appeared as heutagogy "resurfaced as a learning approach after a decade of limited attention", and gives the clearest contrast: in andragogy, curriculum and assessment are "designed by the instructor according to the learner needs", while in heutagogy "the learner sets the learning course". She says the name is "based on the Greek for "self"". She also describes a progression from pedagogy to andragogy to heutagogy as learners mature, and adds that "less mature learners require more instructor guidance and course scaffolding". I read maturity here as maturity as a learner on the task, which is Knowles' spectrum again, not age. Blaschke herself wrote that these traits and the continuum "require further consideration and definition".

The one comparison I found that tested a heutagogy design against a traditional programme found no statistically significant difference in gains; it was not an equivalence trial.

Case study: Heutagogy in teacher certification in Medan, Indonesia

This was a two-group comparison, and the paper does not say how people were placed in each group. Indonesia certifies teachers through a professional education programme tied to government competency standards. With teaching moved online in the pandemic, researchers tested whether a heutagogy design could prepare the 69 early childhood preservice teachers in Medan State University's 2021/2022 cohort at least as well as the traditional face-to-face programme.

• What they did: 36 participants took a 10-week online programme built on heutagogy, with self-determined learning, reflection on how they learned, collaborative work and tools such as mind maps and Google Classroom. The other 33 took the traditional, instructor-directed programme. Both groups sat the same competency test before and after.
• Evidence and result: In Handayani, Peddell and Yeigh's 2023 paper in the Australian Journal of Teacher Education, the heutagogy group's mean rose from 44.50 to 80.39 and the traditional group's from 40.06 to 75.45, a gain of about 35 to 36 points in each, computed from the published means. The difference between the groups' gains was not significant (p = .614). The authors call the approach "at least as viable as the traditional TPE learning", and they report that participants "desired more explanations about the learning content".
• Limitation: The paper does not report how the groups were formed, and the heutagogy group started higher. The heutagogy group was online and the comparison face-to-face, so the method and the mode are mixed together. The national curriculum fixed the content and objectives, so learners could not set their own course, which is the core of heutagogy. It was one cohort at one university, measured with a knowledge test, and the authors designed and evaluated the programme.
• Lesson: The study detected no difference in gains between the two programmes; it did not establish equal effectiveness, and people new to the profession asked for more explanation. Self-determined learning is a choice for learners who can already steer, not an upgrade every adult course should make.

Pedagogy, andragogy or heutagogy: which to use when

For design purposes, I compare the three along a continuum of learner control, and one programme can use all three for different people or at different stages. This table is my synthesis of the sources above, not a ranking, and I found no study comparing all three head to head.

For the dosing alarm procedure, the duty supervisors get a pedagogical route until their supervised shift, and the plant operators an andragogical one. Heutagogy fits neither group for this content, because the procedure and its limits are fixed. It might fit a senior plant operator's own development plan for the following year.

Criticisms of the andragogy and pedagogy comparison

The comparison has drawn criticism since at least 1984. For the evidence on whether andragogy works, see our adult learning theory guide. These criticisms are aimed at the comparison itself:

• It confuses adult and child learning. Hartree's 1984 critique in the International Journal of Lifelong Education called andragogy "a philosophical position rather than a unified theory of adult learning". As other authors summarise her critique, one of its three difficulties is confusion over how adult learning relates to children's learning.
• The terms themselves were disputed. Davenport and Davenport's 1985 chronology of the andragogy debate traces arguments over the philosophy behind it, how to classify it, and the "general utility or value of the term".
• The experiments found no clear winner. Rachal concluded in 1994, as quoted by Holmes and Abington-Cooper in the Journal of Technology Studies, that "the bulk of the experimental and quasi-experimental work done to date suggests an approximate equivalence between andragogical approaches and pedagogical ones on both achievement and learner satisfaction".

Treat the pair as a design lens: two sets of assumptions to check against each group, not a fact about adults or a ranking of methods. That is how Knowles came to see it too.

Using andragogy and pedagogy with an AI course builder

On 9 October 2026 I tested whether Coursebox changes the structure of a course when the learners change, or only its tone. Coursebox is our product, so weigh this account with that in mind. I wrote two briefs for the dosing alarm course that were identical except for the paragraph describing the learners: one for the 8 plant operators, one for the 15 duty supervisors. Both gave the same facts, the five alarm codes and their actions, the plant log fields and the three scenarios, and asked the builder to "decide its structure, order, activities and checks" to suit the learners. Both required a "Before you start" section, a final scenario on the Saturday-night CL-HIGH alarm and a wrap-up called "Who to call and where to log it" as the very last section.

The basics came back as briefed. Both outlines kept the required sections in order and put the wrap-up last.

The structure did not change with the learners. Both outlines followed the same template: an explanation of the alarm screen, a recall quiz of four or five questions, the CL-HIGH scenario, another quiz and the wrap-up. They differed in wording, by about two minutes of estimated duration, and in one extra GAS alarm page for the duty supervisors. The plant operators' course did not start from a task, offered no way to skip and never asked about an alarm they had handled on the old controller, and it still had two recall quizzes on content they knew. The duty supervisors' course had no clearly worked example before the questions and no check they had to pass. The audience description changed the voice, not the assumptions the course made about the learner.

One message per course, naming the assumptions, fixed the outlines. For the operators I asked for the Tuesday CL-LOW scenario before any explanation, a self-check telling them what they could skip, an open-text question about an alarm on the old controller and scenarios in place of the recall quizzes. For the supervisors I asked for a worked CL-HIGH example before any question, a fixed order, a final check they must pass and a note that a plant operator signs them off on a supervised shift. After a reload, the two outlines differed in four structural ways: the opening, choice, use of experience and checks. The operators' outline had no quizzes left, and the supervisors' outline had grown to 10 pages and about 46 minutes, with three quizzes including a 10-question final check.

Then I opened the generated courses, because an outline is not the course. In the supervisors' course, the worked example page came before any question, as briefed: the supervisor "reads the display, closes the pool to swimmers, and begins a calm handoff to the on-call plant operator", and does not "adjust dosing, handle chemicals, or attempt any repair work". The final check requires 80% to pass. I did not verify that every critical safety response must be correct. Before use, check GAS response, pool closure, escalation and authority limits individually; a total score can hide a dangerous answer. In the operators' course, the task-first opening had been replaced at generation by the builder's standard welcome page, so the change that counts most for experienced staff was gone. I rewrote that page in the page editor as "Start here: a Tuesday alarm and a self-check", with the CL-LOW task first and four cards on what operators can skim and what they should not skip, and reloaded it to confirm the change had saved. With that, the operators' course opens on an alarm task and tells them what they can skip, and the supervisors' course walks them through a full response before testing them.

I compared outlines and a few rendered pages, and I did not run either course with plant operators or duty supervisors. The sign-off note did not appear on any generated page, and in both courses the builder replaced our "Who to call and where to log it" wrap-up with its own Key Takeaways page, so both need adding by hand. In this example, the plant operator authorises independent work after the supervised shift; the course result does not supply that permission.

Before you brief a course like this, check where each group sits on the assumptions, because a clear brief gives the builder a better basis for choosing the route, and the generated pages still need checking.

Putting andragogy and pedagogy to work

In Knowles' comparison, pedagogy and andragogy are two sets of assumptions about learners, and the right one depends on the group and the content. The choice between them is not a choice between children and adults. Use andragogical strategies where those assumptions are realistic for this group on this content, and pedagogical ones where learners are new to it, errors are costly, they will act alone soon or the task gives little feedback. Then plan how each learner moves towards more control as they show they can handle it.

Before you build your next course, write one line per learner group in the brief: what they already know about this content, what an error would cost, whether they will act alone and whether the task tells them when they get it wrong. If two groups get different answers, give them two routes.