Why Pedagogy Fundamentals Matter More Than Ever — Lessons from The Essentials of Instructional Design

This is article 10 of 12 in a series on why pedagogy fundamentals matter more than ever in the AI age.

A confession.

About a year ago, I was at my desk on a Sunday night, trying to build a six-week unit on a topic I had taught a dozen times before. For the first time in my career, I felt a small, cold feeling in my chest about my craft.

I felt redundant.

Because I had just watched an AI, in about ninety seconds, produce a six-week unit. With objectives. With assessments. With a rubric. With differentiated worksheets for three reading levels.

It was fine. It wasn't great. It wasn't visibly broken either.

I thought — what exactly do I do now?

What is the part of my job that this thing cannot do?

I'm going to give you the answer in this article. The answer is more reassuring, and more demanding, than you might expect.

The thesis

AI made content production trivial. Instructional design has never been more strategic.

Not less. More.

The part of your craft that the machine cannot do — the analysis, the alignment, the evaluation, the iteration — has just become the most valuable thing you own.

The "prompt-and-publish" problem

There's a movement I want to name.

Call it the "prompt-and-publish" movement. Describe the topic. Hand it to the AI. The AI produces the lesson, the assessment, the rubric. If something feels off, just tweak the prompt and run again.

You don't go back to the analysis.

You don't ask whether you ever did the analysis.

You just regenerate.

I tried this. Honestly. I generated units. I taught them. The lessons "ran." The students completed the activities.

Then I'd give a real assessment two weeks later, and I'd discover I had no idea whether I'd taught what I thought I'd taught — because I'd never written down what I was trying to change in the kid in the first place.

I had skipped the most important step. The AI had skipped it with me.

We had cheerfully co-produced what I can only describe as content that looked like instruction without ever being instruction.

That's vibe-designing. It doesn't survive contact with actual students.

The book

The Essentials of Instructional Design by Abbie Brown and Timothy Green. Routledge, 3rd edition, 2016.

Not flashy. A textbook. One of the most underread books in the entire AI-and-education conversation right now.

It just patiently lays out what the discipline of design has been doing for a hundred years.

Brown and Green define the work:

"Instructional design is the systematic and reflective process of translating principles of learning and instruction into plans for instructional materials, activities, information resources, and evaluation."

The job is not "produce content." The job is translation — from what is known about how people learn, into a plan that includes activities, materials, and evaluation.

The artifact is the smallest part of the work.

The translation is the work.

They sharpen this, citing Reigeluth:

Optimal means. Desired ends.

Not "make the thing." Devise the means. Choose the ends. Design first. Produce after.

This is the part the AI does not do. AI is a content production engine. Structurally, a really fast, really fluent means. It has no opinion about what the desired end is. It has no opinion about whether instruction is even the right intervention. It has no opinion about who your kids are.

It will happily produce a unit on the French Revolution that is technically correct, beautifully formatted, and pedagogically meaningless — because we didn't tell it the end we were trying to achieve, or whether instruction was the right tool for the job.

What Brown & Green actually say

On purpose.

"Instructional design is conducted for a purpose. That purpose is to help produce a desired change."

Not a delivered topic. Not a covered standard. A change in the learner.

"What is the change being requested? Who is being asked to change? Where will this change need to take place? Is instruction the most appropriate means for bringing about the desired change?"

That last question is the one. The default answer of every AI generator is "yes." It will never come back and say, "Actually, this looks like a motivation problem, not a knowledge gap. You don't need a lesson. You need a five-minute conversation."

"Problems that on the surface seem to require an instructional intervention can often be solved with a change in policy, coaching, or the environment."

The AI cannot tell the difference. You have to.

On learners.

"There is not much point to creating an instructional intervention that the intended audience cannot or will not use."

That's the whole problem with default AI output. It is designed for the cannot or will not use learner — the abstract average. The internet doesn't know your kids. The model doesn't know your kids.

It produces for "students" — the ghost in the dataset.

On objectives.

"There is no point in creating any form of instruction without first setting goals for that instruction. The instructional intervention has to be designed to do something — to cause some change in the learner's knowledge, skill, or attitude. Otherwise, it is nothing more than a toy, a collection of artifacts, or an aimless discussion or presentation."

That last clause is a terrifyingly accurate description of half the AI-generated unit plans I have personally produced and shipped.

Brown and Green then quote Mager. The famous A-B-C-D structure:

"Action: identify the action the learner will take when he or she has achieved the objective. Condition: describe the relevant conditions under which the learner will act. Criterion: specify how well the learner must perform the action."

The thing 90% of AI prompts cheerfully skip. We type "objective: photosynthesis" and the AI goes along with it.

On alignment.

"The learner evaluation should be derived from the instructional objectives. There needs to be a direct relationship between the instructional objectives and learner evaluation. If there is, the learner evaluation will be authentic."

Authentic. Not "rigorous." Authentic. The assessment actually measures the thing you said you were going to teach.

On evaluation.

"Evaluation refers to the process for determining the success level of an individual or product based on data and then making decisions based on this success level."

Based on data. Not based on vibes. Not based on what the AI thinks of itself.

"Sustained and varied evaluation is a crucial element of the instructional design process."

Not a nice-to-have. The mechanism by which a design improves.

On iteration.

"The essential idea behind rapid prototyping is to arrive at a final product through the creation of a number of prototypes."

A number of prototypes. The artifact converges over time. It does not arrive whole.

The warning Brown and Green pin in (citing Tripp and Bichelmeyer):

"A design is only a hypothesis."

Every lesson you ship is a guess. The discipline is not in producing it. The discipline is in testing the guess and updating it.

What you do on Monday

Six skills. Each one a deliberate move from artifact-production to design discipline.

1. Diagnose Before You Generate.

Before you open any AI tool, answer four questions in writing — even one sentence each:

• What change am I trying to produce? (In behavior. In skill. In thinking.)
• Who is being asked to change, and what are they currently doing?
• What is causing the gap? (Knowledge gap, skill gap, motivation gap, environment gap.)
• Is instruction the right intervention? (Or is it feedback, policy, coaching, structure, or just time?)

Prompt pattern:

"I am designing for a real change, not covering a topic. The change I want is: [insert]. The learners are currently doing: [insert]. I have diagnosed the gap as primarily a [knowledge / skill / motivation / habit / environment] gap. Before suggesting any instructional materials, push back on whether instruction is actually the right intervention. If it is, propose the smallest viable intervention that would produce the change. If it isn't, tell me what would."

Inviting pushback before generation turns the AI from a vending machine into a thinking partner.

2. Learner Compass.

Before you generate a single AI lesson for a class, write a one-page learner profile.

• Who are these kids? (Specifics: age, context, prior unit, language, school culture.)
• What do they already know? Where are they fluent? Where are they shaky?
• What misconceptions do they bring in every single year?
• What real situations would make this content matter to them?

Paste this profile into every AI conversation about this class. Make the AI design for these kids, not for "students."

3. Objective As Performance.

Before you generate anything, rewrite your objective so it has all four ABCD components:

• Audience — who exactly?
• Behavior — what observable action will they perform?
• Conditions — under what circumstances?
• Degree — to what standard?

If you cannot fill all four, you do not have an objective. You have a topic dressed up.

"Rewrite the following objective in Mager's ABCD form. If the original is missing any of the four — especially an observable behavior — flag it explicitly and propose three alternative observable behaviors I might choose between. Do not produce a lesson until I have approved the rewritten objective."

4. The Aligned Triple.

Before you ship any AI-generated lesson, lay these three things side by side on one page:

• The Objective — written as ABCD performance, with one specific verb.
• The Main Activity — what the student does during the lesson.
• The Assessment — what we use to decide whether they got it.

Do all three ride on the same verb?

If the objective says "analyze," does the activity make them analyze? Does the assessment make them analyze?

"Audit my objective, activity, and assessment for alignment. Identify the cognitive level of the verb in the objective; state the cognitive level the activity actually demands; state the cognitive level the assessment actually measures. If any of the three are at different levels, flag the misalignment and propose changes to bring all three onto the same verb."

Misalignment is the silent killer of AI-built instruction. The artifacts look beautiful. They don't agree with each other.

5. Evaluate the Learning, Not the Lesson.

After every AI-built lesson, before you ship the next one:

• Formative pass — pick three students across the ability range. Look at what they actually produced. Compare it to the objective. Where did the design help? Where did it leak?
• Summative pass — at unit's end: did the change I set out to produce actually occur? Not "did it feel good." Not "did the AI like it."

The AI does not get to grade itself. You grade the learning.

"I am going to paste three real student artifacts produced during this lesson, plus my original objective. Do not evaluate whether the lesson was 'good.' Instead: for each artifact, identify specific evidence (or absence of evidence) that the student demonstrated the behavior named in the objective. Then propose two specific changes to the lesson design — not the prompt, the design — that would close the most common gap you observed."

6. The Living Design.

For every unit, keep a one-page Design Memory alongside the artifacts. After every run, update it.

• Hypothesis — what change in the learner did I think this unit would produce?
• Evidence — what did students actually do?
• Patch — what specific design change am I committing to for next time? (Not "improve engagement." A specific structural change.)
• Open question — what do I still not know?

Save the memory with the unit. Open it before you regenerate next year.

Refuse to start from a blank prompt as long as a design memory exists.

"Below is the unit I taught last year, plus my Design Memory of what I learned. Do not regenerate from scratch. Propose targeted revisions that address each Patch and Open question. Where my memory is silent or vague, ask me a clarifying question rather than guessing."

The shift from regenerate to iterate is the difference between AI as a churn machine and AI as a compounding tool.

The slogan

The Kent Beck refrain again. Invest in the design of the system every day.

The instructional designer's version:

That's the whole job in the AI age.

The part the machine can do — produce content, format text, draft examples, generate worksheets, make the rubric look pretty, write the parent email — is going to keep getting cheaper, faster, better. Forever.

Don't fight it. Don't pretend it's not happening.

The part that is irreducibly you — naming the desired change, knowing the learners, writing the objective as a behavior, aligning the triple, evaluating the learning, iterating on the design — does not get cheaper.

That part is the strategic layer of instructional design.

AI can operate at the tactical layer all day long.

It cannot, structurally cannot, operate at the strategic layer.

That layer requires somebody who has stood in a room with these specific kids and decided what change actually matters.

What's next

Article 11 lands April 2. Cammy Bean's The Accidental Instructional Designer — on why every teacher is now an instructional designer, whether they planned to be or not.

If this series is landing for you, the EDodo flagship — AI-Powered Learning Design — is the cohort version of all this.

Brown and Green have been telling us this for ten years. Mager has been telling us this for fifty.

The discipline already had the answer.

We just stopped reading the books because the new tool felt like a new world.

It isn't a new world. It's the same world, with a much faster artifact-generator. The discipline is still the discipline. More now, not less.

The fundamentals are not the past.

The fundamentals are the strategic layer.

The strategic layer is now where the entire profession lives.

Source: Brown, A. H., & Green, T. D. (2016). The Essentials of Instructional Design (3rd ed.). Routledge. All quotes verbatim from the book.