The Reflective Architect
A Field Guide to Intentional Friction
I have recently discussed the broader dialectic between the algorithmic “Palace” and the human “Doghouse” in my long-form piece, The Palace and the Doghouse. The Palace is the algorithmic system of totalized logic: a smooth, frictionless system that prioritizes a Hegelian sense of the perfections of the infinite achieved through statistical probability and polish. In contrast, the Doghouse is the gritty, limited, and often contradictory reality of the Existing Individual.
Theory is great. But, if we are to move from theory to practice, we must confront the mechanical reality of our tools and what that means for this contrast.
The problem with AI-assisted writing isn’t just that it’s “too smart” or that it tricks us into appearing more competent than we are. Those may be problems, but they’re not the first thing I was thinking of. Instead, what I’m worried about right now is that it’s too smooth.
Usually what I’m hearing are fears about cognitive offloading or the loss of particular skills as a result of AI taking over tasks that have been essential for generations to building particular kinds of metacognitive capacities. I don’t want to take away from the importance of this argument. It’s real, and there are real risks in over-reliance on AI. We may know that the seduction of this smoothness could be bad for us in the long run, but it’s hard to resist the slickness of its efficiency.
And yet what I’m more worried about right now is the way in which this smoothness reveals a kind of Subjective Atrophy. When we use a Large Language Model (LLM), we are interacting with a system that by design erases the “Existing Individual.” It replaces the “View from Here”—the gritty, limited, often contradictory reality of your own existence—with a “View from Nowhere.” Feminist standpoint theory and post-humanism have decried the problem of the view from nowhere for decades. The view from nowhere threatens to homogenize the world and discredit difference. This is the kind of erasure of the existing individual that A.I. has put into our pockets.
So, to instead be a Reflective Architect in the age of AI is to deliberately design a workflow that disrupts this erasure. It is a refusal to let the “Absolute Idea” of the machine smooth over the jagged edges of your own subjectivity. It requires the intentional insertion of friction: a disruptive role that fractures the Hegelian System to make space for the Self.
That is not necessarily easy to do with a tool that by its very design wants to flatter you at every turn. I hardly pretend to have all of the answers, but this is how I would frame a step-by-step protocol for architecting your own “Doghouse-Compatible” creative process.
Step 1: The “View from Here” Inventory
Before you touch the AI, you must anchor your inquiry in the Specific Badness of your initial thoughts. These are the ideas that are half-baked and not pretty. They are the things that we often try to hide in an educational system that values product and polish above all else. The AI, well-trained on decades of corpus writing that reflects this value place on polished product, systematically deletes the “gritty details” of reality because they are statistically surprising. Your job is to find them, because we will use them as boundaries for how the AI will amplify your thinking.
Action: List three “Stains” or “Gritty Details” that are sensory, local, and immediate to your topic. These aren’t ways to AI-proof something, as though AI couldn’t replicate them. But it does narrow the scope of the space in which AI can homogenize your voice.
Example: If writing about “Institutional Identity,” don’t write about “academic excellence” or “unparalleled student support.” Write about the specific, permanent coffee ring on the Dean’s conference table that looks like a map of a ghost town.
Step 2: The “Friction-Prone” Prompt Sequence
The goal is to use the AI to heighten the tension of your ideas, not to resolve it. In a Hegelian system, tension is a problem to be solved—a contradiction to be synthesized into a higher, smoother unity. But for the Existing Individual, tension is the site of the Qualitative Leap. If you allow the AI to resolve the tension for you, you are abdicating the very choice that constitutes your Self.
We must recognize that there are different ways the AI attempts to resolve this tension: it might offer a middle-of-the-road compromise, it might ignore the “gritty details” entirely to preserve its internal logic, or it might flip-flop between incompatible views without actually standing anywhere. By structuring our prompts into the following stages, we force the AI to hold the tension open, exposing the gaps where only a human choice can bridge the distance.
Try this three-stage prompting structure:
Stage A: The Generic Contrast
“I am writing about [Topic]. Provide the most probable, ‘smooth’ consensus view on this subject. Identify the three most common metaphors the average person uses to describe it.”
Rationale: This stage isolates the “Palace” as an external object. By forcing the AI to articulate the “smooth consensus,” you create the distance necessary to see its logic as a “View from Nowhere” rather than your own voice. Essentially, you are mapping the boundaries of the algorithmic system so you can eventually step outside them. You are pressing the LLM to do the wrong thing so that it has a straw man to knock down.
Stage B: The Intra-Action (The Wound)
“Now, contrast those consensus views with these specific ‘stains’: [Your Details from Step 1]. How do these gritty details ‘wound’ the smooth logic of these metaphors Where do my details break down the consensus view?”
Rationale: In this stage, you force the “Absolute Idea” of the AI to collide with your locally-anchored “Stains.” This is where things to fracture. The “View from Nowhere” cannot remain objective when it is forced to account for a specific coffee ring on a specific table. You are re-introducing the friction of the real into a system designed to ignore it.
I think it’s really important to remember that Stage B is very often a dialogical process and not a one-off turn in the chat. An LLM is designed to be helpful and perfect. It’s not going to give up that end easily. In many instances, at stage B, I’ve watched an LLM completely flip on its head and simply throw out the consensus views, acknowledging how brilliant I was for pointing out why it was wrong.
This kind of flattery isn’t the goal. The goal is to get it to contrast the consensus view with what is distinct to you and your place in the world. To make this work, you have to push the AI, and in pushing the AI, you have to push yourself to be clear about how to prevent it from letting go of the consensus view too quickly. This stage is an art form that teachers are well-suited to. It looks a lot like the process a teacher might go through in order to get a student to clarify an idea.
Stage C: The Refusal of Synthesis
“Identify the core contradiction between the general probability and my local reality. Propose two incompatible responses. Do NOT synthesize them.”
Rationale: This stage is perhaps the most difficult for the LLM and your prompting of it. Many of the things that we make fun of as being textbook instances of AI writing are versions of a bad dialectic that the AI is trying to reproduce as a very typical mode of human thinking. All of the “em-dashes and “Not this..., but that...” statements that we see AI produce are reflections of this bad dialectic. They are dialectics that tend to reduce to the middle, trying to find an average between the polarity that the AI has set up, or to set up the polarity as a false binary that collapses to only one side of the available positions. Everything becomes a binary.
If stages A and B have gone well, some of this will have been nipped in the bud, but there will still be a tendency to do a slightly more sophisticated form of dialectical synthesis (a Hegelian move toward overturning both A and B for some new idea C that includes both of them). This response is sneaky because it feels like it might be your own voice, even when it’s not. Moreover, many of our standard AI workflows often ask the tool to “bring it all together” or “summarize the key points,” which invites this synthesis that is a form of subjective erasure.
By explicitly forbidding synthesis and demanding incompatible paths, you preserve the “Dizziness of Freedom” that only a human choice can bridge. You are reclaiming the authority to choose, which is the only way to proceed.
Step 3 (Optional): Architecting the Evaluation (The Manager’s Protocol)
By the time you complete Stage C of the prompting sequence, you have achieved something the AI cannot: The Clarity of Tension. You now have the raw materials—the “stains” and the incompatible paths—to move into the actual construction of your argument. If you choose to use the AI for the drafting phase at this point, you will find it much harder for the machine to eliminate your voice, because you have already anchored the creative space in your specific, local reality.
However, this protocol often fails not because the architect is lazy, but because the Bureaucracy demands polish over process. If you are a manager or educator, this step is for you.
As I argued in The Un-cheatable Assignment, we must recognize the “technical futility” of trying to detect AI-generated products. If you grade the artifact (the Palace), your people will simply rent that Palace from the AI. To protect the “Doghouse,” you must shift your focus from the Product to the Process.
Think of this as a “Pre-flight Checklist”: a model drawn from Specifications Grading that prioritizes clear standards of presence over subjective counts of percentage points. I am not interested in whether or how the AI was used; I am looking for the evidence of an Existing Individual.
The Checklist of Care (Evaluation of Process)
Voice Density: Does the writing reflect a specific, non-averageable voice? Is there a cadence or a set of choices that feels like a confession of growth rather than a manipulation of signs?
Local Gravity: Does the text feel rooted in a specific location or context (the “Stains”) that probability could never predict? Is the “coffee ring on the Dean’s table” present, or has it been smoothed into “academic excellence”?
The Dizziness Check: Does the author articulate the difficulty of having to choose between incompatible paths? Do you see the “Qualitative Leap” of a radical individual or just a “smooth” Hegelian synthesis?
This is one of the hard parts I think for many faculty in particular who are grading. We want to look at these three points and say...”Kind of...” We want to read charitably of the student and suggest that we can encourage them to do more. But we want to treat this act in specifications grading as a simple binary: zero or one. You have to decide how strictly you want to evaluate that. But this isn’t a 80% there sort of scenario. If the work doesn’t meet these specifications, it is a “View from Nowhere”—an empty Palace. Send it back for iteration. Iteration is good! This is the work of revision that is so critical to sharpening our thinking. The goal is to reward the inhabitant, not the architect of the void.
Is there sufficient voice density? Yes or no?
Is there local gravity? Yes or no?
Does it pass the dizziness check? Yes or no?
It is a point where being severe is to the benefit of the student being graded, or the essay that we are reading, or the work we’re evaluating. Our job is to make that stronger.
The Quantitative Advantage
By choosing the “Doghouse” over the Palace, you aren’t just being “ethical.” You are being Original. In an economy of infinite synthesis, the only surviving credential is the Qualitative Leap as the moment you stake your voice on a choice that probability could never predict.
Build the Doghouse. Inhabit the friction. It’s the only place where the Self can actually live.
🔍 Key Entities & Context
The Palace vs. The Doghouse: A dialectic framework for AI adoption, contrasting totalized algorithmic logic with the lived experience of the existing individual.
Reflective Architect: A proposed professional persona defining an intentional workflow that re-introduces human friction into AI-assisted creative processes.
Subjective Atrophy: The erosion of individual agency and local context (”The View from Here”) caused by over-reliance on “smooth” AI-generated outputs (”The View from Nowhere”).
The Qualitative Leap: A Kierkegaardian concept applied to AI, defining the non-probabilistic moment of human choice that constitutes the “Last Human Credential.”
Institutional Context: Focus on Higher Education identity, Specifications Grading (Linda Nilson), and Process-Based Assessment as a defense against algorithmically-driven product fetishism.