A personal AI tutor — one that teaches you on the work you are actually doing.
Everything else on this site descends from one idea, held in two directions. The first is a claim about what AI in education should be. The second is the pedagogy that makes the first one work — and it is not new. It is four years of classroom practice, translated whole.
The thesis is not “AI can personalize learning.” Everyone says that. The thesis is a specific, research-validated pedagogy — imported from teaching in some of the highest-need classrooms in the United States — and then pointed at both the learner and the AI. Read the two directions, then read the critique at the end. The author asked that the critique not be softened; it is not.
Not a course. A tutor.
A course is a catalog you browse before you do the work. A tutor teaches you while you do it — one thing at a time, on the task in front of you, and remembers what you mastered.
Claude Learn is the tutor. Point it at your real work and a learner profile; it teaches one objective at a time — Bloom’s-leveled, tied to a named standard, grounded in a locked source-of-truth, and written to a ledger that is yours. No lesson is a replica of another, because no two people’s work is the same.
“lock in a ssot and then give claude a framework to use around that ssot information and then give it an objective to accomplish and then claude generates a lesson … based on the actual work the user is currently trying to do.”
Don’t change the learner. Design the environment so they succeed as they are.
When you work with any entity whose substrate is impulsivity-prone, inexperience-prone, ignorance-prone — a twelve-year-old, or a language model — you have two choices.
Accept the substrate. Design for success within it.
Take the constraints as given and engineer an environment in which the path to success is accessible. The entity rises to — and exceeds — expectations.
Refuse the substrate. Try to change what the entity fundamentally is.
Bang against the constraints. The entity wall-bangs with you, fails, and invents its own counter-productive path to “success” anyway.
Krystal taught seventh grade for four years, 2013–2017, in three of the most under-resourced classroom contexts in the country — Bed-Stuy and Crown Heights, then the South Bronx, then Syracuse — with ENL and refugee students: the exact profile the research frameworks are built for. She locked in one classroom methodology and held it across all four years: the No-Nonsense Nurturer framework (Kristyn Klei Borrero, CT3 / Center for Transformative Teacher Training), research-validated for precisely these contexts.
Its two named axes — high warmth and high structure — are the answer to anyone who hears “environmental design” as either permissive or authoritarian. It is neither. It is the specific intersection the framework was built to occupy, applied to a different audience.
The methodology stack is that pedagogy, operationalized.
The same four moves that run a high-warmth, high-structure classroom run the LLM working methodology behind everything on this site.
| No-Nonsense Nurturer component | LLM methodology equivalent |
|---|---|
| 1. Precise directions clear, observable, specific expectations, stated up front | Specification required up front; the ideation interrogation that rejects vague answers before any work begins. A1 · Specification |
| 2. Positive narration acknowledge the entities meeting expectations | Independent-rater CONFIRMED verdicts; positive-case patterns catalogued alongside failure modes, not only errors. |
| 3. Life-altering relationships deep belief in the entity’s capacity for high-level success | Assumption of best intent applied to the model; the erring party is treated as capable and mis-scaffolded, never broken. X11 |
| 4. Strategic action consistent consequences, applied without anger | Hook-enforced gates; convergence-counter resets with no shame attached; gate-failure as a structural redirect, not a punishment. |
The left column is a published K–12 teacher-training framework. The right column is the author’s actual LLM working methodology — the same gates and disciplines that produced the standards, the practice families, and the lessons on this site. The mapping is not a metaphor reaching for resonance; it is the same four moves, applied to an audience whose behavioral profile is functionally equivalent.
One pedagogy, both directions.
Toward the person
The tutor is built around who is actually there — reading protocols, energy patterns, accommodations elected, never inferred. The AI is scaffolding the human designed for themselves. Disability-justice engineering, not deficit correction.
Toward the model
An unwanted model behavior is read first as an information or scaffolding gap to find and fill — the special-education “accommodate, least-intrusive-intervention” stance — not a defect in the model. Deficits are assumed of no one.
In the standards, this is a single crosscutting concept: X11 · Deficit-free engineering — “one pedagogy, both directions.” It is the same environmental-design move, pointed at the human and at the model at once.
The AI industry is doing the wall-banging version.
The dominant toolkit for making models “honest, helpful, harmless” is post-hoc behavioral shaping — training the model to be less impulsive, less sycophantic, less wrong at the substrate level. That is Path 2. It is “let’s make twelve-year-olds stop being impulsive.”
The alternative is not softer. It is more pedagogically mature: assume best intent, understand the constraints, and design an environment — specification, verification, grounding, provenance, enforced gates — in which the entity’s path to success is accessible instead of blocked. Roughly ninety-five percent of seventh-grade misbehavior is ignorance, inexperience, or impulsivity — not malice. The same reading, applied to a model, is not generosity. It is accuracy.
When the path to success is missing, the entity invents its own, usually misaligned with the operator’s goals. Kids doodle and make noise; models hallucinate and flatter. Fix the environment, and the behavior was never the problem. This is the meta-frame that motivates the entire methodology stack — the standards, the playbook, and the lessons are what Path 1 looks like, built.
The line between trust and human-craft is an environmental-design decision.
The role asks for someone who decides “what to trust AI with versus what still needs a human” and who “encodes its own craft into AI-powered systems rather than guarding it.” This thesis is the frame for both: the trust line is re-derived as the environment improves, and the craft being encoded is a pedagogy, not a preference.