Playbook: Alex for Theory Development & Writing

Your reference for applying AI to constructing, testing, refining, and publishing theoretical contributions, from initial conjecture to peer-reviewed publication. Ready-to-run prompts built around the discipline of theory construction, not the illusion that AI can theorize for you.

What This Guide Is Not

This is not a habit formation guide (see Self-Study Guide for that). This is a domain use-case library: the specific ways dialog engineering supports rigorous theory development at every stage, from initial observation to published framework.

Where to Practice These Prompts

Every prompt in this guide works with any AI assistant: ChatGPT, Claude, GitHub Copilot, Gemini, or whatever tool you prefer. The prompts are the skill; the tool is just where you type them. Pick the one you’re comfortable with and start today.

For an integrated experience, the Alex VS Code extension (free) was purpose-built for this workshop. It understands research context, lets you save effective prompts with /saveinsight, and brings your playbook and practice exercises into one workspace. VS Code is a free editor that takes minutes to set up, even if you’ve never used it before.

You don’t need a specific tool to benefit. You need the discipline of treating AI as a thinking partner for the hardest intellectual work, not just a drafting assistant.

Core Principle for Theory Development

Theory development is where AI assistance is most dangerous and most valuable in the same breath. The danger: AI generates fluent, confident prose that sounds like theoretical contribution but adds nothing: repackaging existing ideas in new vocabulary, inflating observations into “frameworks,” or producing unfalsifiable propositions dressed as theory. The value: AI can systematically pressure-test your logic, surface competing explanations, identify boundary conditions, and help you articulate what your theory actually predicts that existing theories do not.

The theorist’s job is irreplaceable: observing phenomena, recognizing patterns, making the creative leap to a mechanism or explanation that nobody has proposed before. AI cannot do this. What AI can do is everything that surrounds that leap: organizing existing literature, stress-testing propositions, mapping construct relationships, identifying where your logic is circular, and helping you write with the precision that reviewers demand.

Use AI to sharpen the theory. Never let it generate the theory for you.

The CSAR Loop in Theory Work

Every theory development session benefits from the four-phase cycle: Clarify → Summarize → Act → Reflect.

The Reflect phase is especially critical in theory work. First drafts of theoretical arguments almost always contain hidden circularity, where the conclusion is restated as a premise in different words. The CSAR loop catches this before a reviewer does.

The Seven Use Cases

1. Phenomenon Articulation — Defining What Needs Explaining

The theory challenge: Every theory begins with a phenomenon that existing explanations fail to account for. The most common mistake in early-stage theory work is not that the phenomenon is uninteresting; it is that it is too vaguely stated to be theoretically productive. “Organizations struggle with AI adoption” is an observation, not a phenomenon statement. A phenomenon must be specific enough that two researchers could agree on whether they are observing the same thing.

Prompt pattern:

I'm a [role] developing theory in [field].
I've observed this phenomenon: [describe what you're seeing in concrete terms].
Current explanations from the literature: [list 2-3 existing theories that partially address it].
What they fail to explain: [what remains unexplained or poorly explained].

Help me:
1. Sharpen this into a phenomenon statement specific enough for a theory paper
2. Identify whether what I'm calling "unexplained" is genuinely unexplained or just under-studied
3. Suggest how I might bound this phenomenon so it's tractable for a single paper
4. Flag any cases where existing theory might already explain this and I'm overstating the gap

Follow-up prompts:

My phenomenon statement currently reads: [paste]. A hostile reviewer would say existing theory already explains this because [X]. How do I pre-empt that objection?

I have three related observations. Should these be treated as one phenomenon or three? What are the criteria for deciding?

What is the minimal empirical evidence I need to demonstrate this phenomenon is real (not just anecdotal) before I build theory around it?

Try this now: Take something you’ve noticed in your research domain that current theory doesn’t fully explain. Write a two-sentence phenomenon statement and ask the AI to evaluate whether it’s specific enough to anchor a theoretical contribution.

2. Construct Development — Building the Vocabulary

The theory challenge: A construct is an abstraction that names something real but not directly measurable: “psychological safety,” “absorptive capacity,” “institutional logic.” Good theory requires precise constructs with clear definitions, clean boundaries from adjacent constructs, and explicit dimensionality. The most common failure: borrowing a construct from another field without adapting its definition, then treating the label as if it were self-explanatory.

Prompt pattern:

I am developing a new construct (or refining an existing one) for my theory of [topic].

The construct I'm working with: [name]
My current definition: [your working definition]
What it captures: [the essence of what you're trying to name]
Adjacent constructs it must be distinguished from: [list similar terms]

Help me:
1. Evaluate whether my definition is precise enough that two researchers could independently identify instances of this construct
2. Identify where my construct overlaps with [adjacent construct] and how to draw a clean boundary
3. Suggest whether this construct is better treated as unidimensional or multidimensional
4. Flag any existing constructs in the literature that already capture what I'm describing

Follow-up prompts:

How would I operationalize this construct for empirical testing? What observable indicators would distinguish high from low levels?

I've defined this construct at the individual level. Does it also apply at the team or organizational level, and would the definition need to change?

What is the simplest example I can give that makes this construct intuitively clear to a reviewer outside my subfield?

Try this now: Pick one construct central to your current theorizing. Write out its definition. Then ask the AI what overlapping constructs a reviewer might confuse it with, and whether your definition draws a clear enough boundary.

3. Proposition Logic — From “I Think” to “Therefore”

The theory challenge: A proposition is a statement about how constructs relate to each other. “Trust influences adoption” is not a proposition; it is a direction. “Trust reduces the perceived risk of capability loss, which in turn increases the willingness to delegate high-stakes tasks to AI” is a proposition; it names the mechanism. Theory papers live or die at the mechanism level. Reviewers will accept your phenomenon and your constructs; they will attack your mechanism.

Prompt pattern:

I am developing propositions for my theory of [topic].

Construct A: [name and definition]
Construct B: [name and definition]
My proposed relationship: [A influences B through mechanism M]

Help me:
1. Evaluate whether the mechanism I've proposed is actually a mechanism (explains HOW) or just a restatement of the relationship
2. Identify the hidden assumptions this mechanism requires to work
3. Suggest boundary conditions — under what circumstances would this mechanism NOT operate?
4. Check whether the logic is circular (does Construct B's definition already contain Construct A?)

Follow-up prompts:

I have five propositions. Read them together. Are they internally consistent, or does Proposition 3 contradict the logic behind Proposition 1?

What is the strongest alternative mechanism that could explain the same relationship? How does my mechanism differ in its predictions?

This proposition is currently stated in one direction (A → B). Under what conditions might the relationship reverse?

Try this now: Take one proposition from your current theory work. Isolate the mechanism, the HOW. Ask the AI: “Is this a genuine mechanism, or am I just restating the relationship using different words?” The answer is often uncomfortable and always useful.

4. Boundary Conditions — Where Theory Stops Working

The theory challenge: No theory is universal. Stating boundary conditions (the conditions under which your theory applies and where it breaks down) is what separates a theoretical contribution from an overgeneralized claim. Reviewers are trained to ask “when does this not hold?” The theorist who answers that question proactively earns credibility. The one who dodges it invites rejection.

Prompt pattern:

My theory proposes that [core claim].
It is developed in the context of [specific domain, setting, or population].
The key constructs are: [list].
The mechanism is: [brief description].

Help me:
1. Identify the implicit boundary conditions I haven't stated — what must be true about the context for this theory to apply?
2. Suggest settings or populations where this mechanism would likely fail — and why
3. Evaluate whether my boundary conditions are too narrow (limiting generalizability) or too broad (inviting disconfirmation)
4. Propose one boundary condition that, if tested, would be the strongest test of the theory's limits

Follow-up prompts:

A reviewer says my theory only works in Western, WEIRD contexts. How do I address that without either dismissing the concern or over-qualifying into irrelevance?

I originally developed this for [context A]. A colleague says it should also apply to [context B]. What would need to change (or not change) in the theory for that extension to hold?

What boundary condition would most surprise the field if my theory held across it?

Try this now: State your theory’s domain of application in one sentence. Then ask the AI to name three contexts where it would probably fail. If you cannot explain why it would fail in those contexts, you may not fully understand your own mechanism.

5. Competing Explanations — The Hostile Review Simulation

The theory challenge: Every theory competes with alternatives. A theory paper that ignores competing explanations appears naive; one that dispatches them too easily appears dishonest. The goal is to name the most threatening alternative explanation, demonstrate that you understand it, and show, through logic rather than rhetoric, why your explanation adds something the alternative does not.

Prompt pattern:

My theory proposes that [phenomenon] is explained by [your mechanism].
The strongest existing explanation comes from [competing theory/author]: [brief summary].

Play the role of a reviewer who is deeply committed to the competing explanation.
1. What is the strongest case that the competing theory already explains the phenomenon?
2. Where does the competing theory's explanation fall short — what specific aspect of the phenomenon does it fail to account for?
3. Is there a way to integrate both explanations, or are they genuinely in tension?
4. What empirical test would distinguish between the two explanations?

Follow-up prompts:

I've dismissed the competing explanation with this argument: [paste]. Is this a genuine refutation or am I straw-manning?

What would an advocate of [competing theory] say is the weakest link in MY mechanism?

A reviewer claims my theory is "merely an extension" of [existing theory], not a new contribution. How do I respond?

Try this now: Name the theory that most closely competes with yours. Write two sentences explaining why yours adds something new. Ask the AI to argue the opposite: that yours doesn’t add anything. The quality of your response to that challenge is the quality of your contribution.

6. Writing the Theoretical Argument — Precision Without Jargon

The theory challenge: Writing theory is not the same as writing empirical findings. Empirical sections have a natural structure: method, results, interpretation. Theory sections must construct an argument where each paragraph follows logically from the previous one, building toward propositions that feel both surprising and inevitable. The prose must be precise without being impenetrable, and it must cite without becoming a literature dump.

Prompt pattern:

I am writing the theoretical development section of a paper for [journal/audience].
The argument I need to construct: [brief outline — from starting conditions through mechanism to propositions].
The key challenge in the writing: [where you get stuck — transitions? precision? conciseness? justifying assumptions?].

I will draft sections. For each draft, I need you to:
1. Check whether each paragraph follows logically from the previous one
2. Flag anywhere the argument leaps without justification
3. Identify jargon that assumes shared understanding the reader may not have
4. Mark sentences where I am asserting something that needs either a citation or an explicit assumption statement

Follow-up prompts:

This paragraph is currently 200 words. Where is the fat? What can I cut without losing the argument?

I've used the word "[construct name]" six times in two paragraphs. Help me vary the language without introducing ambiguity.

Read this section as a reviewer in [adjacent field]. What would you find unclear?

This transition between the literature review and the theoretical development is clunky. I need it to signal clearly: "Existing work establishes X. What remains unexplained is Y. The theory I develop addresses Y through Z." Help me write that bridge.

Try this now: Take your weakest paragraph, the one where you know the logic isn’t tight yet. Paste it and ask the AI to mark every point where a skeptical reader would say “why should I believe this?“

7. Visual Theory Representation — Making the Framework Visible

The theory challenge: A theoretical model diagram is not a decoration. It is an argument in visual form, showing which constructs connect, which mechanisms transmit effects, where moderators intervene, and what the boundary conditions exclude. A weak diagram is a box-and-arrow chart that restates the propositions without adding clarity. A strong diagram reveals the structure of the theory at a glance, making relationships and feedback loops visible that prose conceals.

Prompt pattern:

My theory has the following constructs and relationships:
[List constructs and their proposed relationships, including direction and mechanisms]

Moderators: [list conditions that strengthen or weaken relationships]
Boundary conditions: [what is excluded]

Help me:
1. Design a theoretical model diagram that shows the causal logic clearly
2. Decide whether this is best represented as a linear path model, a feedback loop, a multi-level model, or a process model
3. Identify relationships I've described in my propositions but haven't included in the visual
4. Suggest where the visual reveals redundancy or circularity I haven't noticed in the prose

Follow-up prompts:

Create a Mermaid diagram of this theoretical model using simple, clean styling.

This model has nine constructs. That's too many for one figure. How should I decompose it into sub-models? What is the organizing principle?

I have a feedback loop in my model (A → B → C → A). How do I represent this without the diagram looking circular or confusing?

Try this now: Sketch your theoretical model as a list of “X → Y via Z” statements. Ask the AI to convert it into a diagram. The visual will almost certainly reveal a relationship you forgot to include or one you included twice under different names.

The Five Anti-Patterns in Theory Development

Prompt pattern:

Here is my current theoretical argument: [paste key section].
Evaluate it against the five theory anti-patterns:
1. Am I just renaming existing constructs (Vocabulary Swap)?
2. Does any mechanism restate its own conclusion (Circular Mechanism)?
3. Can each proposition be falsified with a realistic empirical test (Unfalsifiable Claim)?
4. Is every construct necessary — which could I remove without losing explanatory power (Kitchen Sink)?
5. Does every paragraph advance the argument or just report literature (Literature Tour)?

Power Moves for Theory Development

Practice Plan

Days 1-5: One Skill Per Day

Months 2-3: Integration

• Week 1-2: Use the CSAR loop for every theory writing session: Clarify your argument before drafting, Summarize the logic, Act on one section, Reflect on where the argument is weakest
• Week 3-4: Run the full anti-pattern check against your current manuscript draft
• Month 3: Present your theory to a colleague and use the AI to pressure-test their objections before revising

The goal is not to produce theory faster. The goal is to produce theory with fewer hidden flaws, so that the peer review process improves the theory instead of exposing its foundations.

Quick Reference

The Seven Use Cases

The Five Anti-Patterns

With the Alex Extension

If you use the Alex VS Code extension (free), these additional capabilities enhance your theory development practice:

Getting started with Alex:

1. Install VS Code → Install GitHub Copilot (free tier works) → Install “Alex Cognitive Architecture”
2. Press Ctrl+Shift+P → “Alex: Initialize Architecture”
3. Open Copilot Chat → Select Alex as the agent
4. Introduce yourself: Hello! My name is [name]. I'm a [role] developing theory in [field].

For the full setup guide, see The Extension.