Foundations: enrolling now. Self-paced. $299. Get on the list.
Foundations
Self-paced · 3 modules · $299
Your prompts keep getting longer and your output keeps getting worse. This course teaches you the structural fix and how to keep your intent intact from start to finish.
You already know how to read a room: a client, a brief, an audience. Figuring out what something needs before you start is how you do your best work. This course teaches you to do the same thing with AI.
Written lessons, applied exercises on your own project, and weekly open office hours. Three modules, gradual release: demonstration, guided practice, independent practice. You finish with a working method built on your own real work.
Module 1
Task Decomposition
Demo, guided practice, independent practice
Your prompts keep getting longer and your output keeps getting worse. The problem is structural: compound instructions degrade in AI the same way they degrade in a classroom. You learn to run a room of specialists. Each one goes deeper than a compound prompt because its attention isn't split across five objectives at once. The AI gives you raw material at depth. You provide the synthesis: the calls the model cannot make because it doesn't have your judgment. You're directing.
Deliverable: A decomposed workflow for one real project, with before/after comparison.
Module 2
Input Inversion
Demo, guided practice, independent practice
Structured prompts compress what makes your thinking yours. Unstructured input (talking, brainstorming, arguing with yourself) contains richer raw material. You learn to dump your thinking and let the tool find the structure. The ideas that emerge are yours because the raw material was yours.
Deliverable: An input practice for one real project, with a recorded brain dump and its structured output.
Module 3
Voice Governance
Demo, guided practice, independent practice
All AI copy sounds the same because voice constraints are applied after generation, not during. You learn to extract how you actually talk (from conversation, not published writing), codify it as generation rules, and apply those rules before the first word is produced. The output sounds like you because the constraints occupied the space where the AI's defaults would have gone.
Deliverable: A voice protocol extracted from your own speech patterns, tested against real output.
What the difference looks like.
Brand design — task decomposition
Same brand. Same model. One compound prompt covering five dimensions at once. Five decomposed prompts, one per dimension. Here is what the logo direction came back as in each version.
Compound Output
Logo Direction
"A wordmark-forward logo using a hand-drawn or letterpress-style treatment with a supporting mark (possibly a stylized pepper or farm element). The mark should feel made by hand, not designed in software. Avoid flames, skulls, cartoons."
Decomposed Output
Logo Direction
"The mark should feel earned. It should carry the weight of someone who has done physical work to produce the thing in the jar. Drawn by someone with dirt under their fingernails, not in software. Reference sign painters' corner initials and printer's marks from the 1500s: compact enough to stamp, distinct enough to identify the shop. Reproducible by hand on a napkin. Must work at bottle-cap size and at eight-foot market-stall scale."
The compound version described what the logo looks like. The decomposed version described what it feels like, what it references, what it must avoid, and what constraints govern every scale. The practitioner now has something to act on, not just a spec to execute.
Essay writing — input inversion
Same writer. Same argument. One structured prompt asking for a complete essay on the topic. One raw input dump (voice notes on the argument, unedited) fed back as the source. Here is what the opening paragraph came back as in each version.
Structured Prompt
Essay Opening
"Distraction is one of the defining challenges of modern work. With the rise of digital technology and constant connectivity, knowledge workers face an unprecedented number of interruptions. Research shows that it takes an average of 23 minutes to regain focus after an interruption. This essay examines the nature of distraction, its costs, and strategies for managing it in the contemporary workplace."
Input Inversion
Essay Opening
"The thing I can't get past is that every productivity system I've tried assumes the problem is time management. It isn't. The problem is that I've built a life where the highest-value work requires a kind of sustained attention I can't sustain anymore. Not because I'm broken — because I've trained myself out of it. The question isn't how to block distractions. It's whether I've become someone who can still do the work."
The first version describes a social problem at category scale. The second names a specific realization from a specific person about their own situation: the actual argument. The input dump contained the real thinking. A structured prompt asked for a topic; the model gave back a taxonomy. An unstructured dump gave the model the writer's actual observations; the model gave back the argument that was already in the raw material.
Fiction writing — voice governance
Same character. Same model. One version with no voice constraints. One version with a protocol extracted from the author's own speech patterns, applied before the first sentence. Here is what the opening came back as in each version.
No Voice Protocol
Opening Scene
"Henka had lived in New City for as long as she could remember. The city was a maze of rusted machinery and broken concrete, walls of salvaged metal that blocked out most of the sky. She moved through the shadows carefully, watching for the corporate drones that swept the sector every three hours. If they spotted her, there would be questions. Questions she couldn't afford."
Voice Protocol Applied
Opening Scene
"Oil and standing water, the smell of both. She pressed into the gap between the housing units and waited. Her shoulder against the corrugated metal, she felt the hum before she heard it. Distant. East. She counted beats without counting — it was just a rhythm her body kept. When the hum faded she moved. The pile was still there."
The first version describes a character in a world. The second renders a world through a body. The difference is not better writing; it is a specific set of constraints applied before the first sentence: sensory priority (smell before sight), emotion as physical state, not internal monologue, short sentences during tension, no exposition. The author's voice held because it was encoded first. The AI's defaults occupied none of the space.
Building a practice — task decomposition
Same practitioner. Same model. One compound prompt asking for a complete business and marketing plan. One focused question asking about positioning only. Here is what came back.
Compound Output
Practice Direction
"1. Define your target market: Conduct market research to identify your ideal clients, their industries, needs, and pain points. Create buyer personas. 2. Develop a brand identity: Create a logo, website, and social media profiles consistent with your brand. 3. Create a portfolio: Showcase your best work. 4. Determine your pricing structure: Research industry standards to ensure your prices are competitive. 5. Create a marketing strategy: Use social media, email marketing, and networking to reach your target market..."
Decomposed Output
Positioning Only
"You spent eight years as a journalist before you started writing for SaaS companies. That is not a career pivot. That is why you know how to find the human story inside a technical product before you write a single sentence about it. Most copywriters learn to write clearly. You learned to find what is true first. There is a specific kind of client — the one whose product does something real but whose copy sounds like everyone else's — who needs exactly what you have. The marketing plan follows from that. The portfolio is evidence of it. Everything else is execution."
The compound version gave five steps that apply equally to any practitioner who has ever read a business blog. The decomposed version, focused on positioning only, named a specific capability built over eight specific years and the exact client who needs it. The difference is not the model; the model had the same information both times. The compound prompt asked it to cover everything at once. The focused prompt left it nowhere to go except into the specifics you gave it.
Campaign copy — voice governance
Same brand. Same model. One prompt asking for email subject lines and opening copy for a product launch. One version with a voice protocol extracted from the brand's existing content. Here is what the copy came back as in each version.
No Voice Protocol
Launch Email
Subject: "Introducing Our New Collection: Made for the Way You Live"
"We're excited to share something we've been working on for a long time. Our new collection was designed with you in mind — crafted from sustainable materials, built to last, and priced for real life. Because you deserve products that work as hard as you do."
Voice Protocol Applied
Launch Email
Subject: "It took us three years. Here's what changed."
"We scrapped the first version in 2022. The materials were right. The weight was wrong. This one is different — not because we found better words for it, but because we fixed what we got wrong the first time."
The first version uses phrases — "crafted," "built to last," "work as hard as you do" — that appear in thousands of brand emails. The model filled the available space with its defaults. The second carries the brand's actual voice because the protocol set constraints before generation: no aspirational language, no "you deserve" framing, product claims grounded in specific development history, short declarative sentences. The AI's defaults had nowhere to go.
The methodology came from a fourth-grade classroom in Sunset Park, Brooklyn. Twelve students, twelve individualized education plans. A compound instruction ("take out your notebook, turn to page 42, answer questions three through seven") is three tasks arriving simultaneously. A student with processing delays hears the first one and starts. The second and third are gone.
The accommodation is decomposition: one instruction at a time, each one specific, each one completable before the next arrives. In that classroom, designing around the processing reality of the receiving system is federal law.
I transferred the framework to AI in 2022 and the fix was identical. A large language model processing a compound prompt distributes attention across all objectives at once. Each gets a fraction. Decompose the prompt and the model gives you depth it couldn't reach when its attention was split five ways.
That transfer is why the course has a name: Accommodation Design. The method predates AI by eight years. It was designed around a different receiving system entirely.
What's included
Gradual release pedagogy: demonstration, guided practice, independent practice (from IDEA-governed special education method)
Applied exercises on your own projects, not fictional examples
Three deliverables you use the day after
Frameworks, templates, and reference materials
Weekly open office hours. Bring your work or your questions. I work through real submissions from the group so you see the method applied to work like yours
Grounded in six published whitepapers with DOIs (citations include Sweller, Cowan, Murdock, Liu et al., Wong et al., Bock, Pennebaker, Flower & Hayes)
Built by a trained educator (Brooklyn College M.A. program, NYC Department of Education, Kingsborough Community College) using the same pedagogical framework the course teaches. The course structure itself demonstrates Accommodation Design. Read the research.
Foundations is complete on its own. For practitioners who want to go deeper, Advanced AI Practice courses (Input Inversion, Lens Extraction, Voice Governance, Semantic Hierarchy, and Coordinator Building) each deepen a single practice on your own real work. See all offerings.
$299
Three modules. Three deliverables. One methodology. Personal review from the founder.
After Foundations, Advanced Practice deepens the methodology and specialist tracks apply it to specific disciplines. Both are future offerings for graduates.
Get on the list.
Self-paced. $299. Personal review from the founder.