In less than five hours, I wrote a textbook and course handbook with AI … and both are good
Source: AI Goes to College
Author: Craig Van Slyke
Original source: https://aigoestocollege.substack.com/p/in-less-than-five-hours-i-wrote-a
Published: 2026-07-14
Source type: essay
Private backup: the full article text is archived in the private repository at archives/articles/aigoestocollege-substack-com-in-less-than-five-hours-i-wrote-a.source.md. It is not published on the public Quartz site.
Summary
Craig Van Slyke describes two fast, hands-on experiments with Codex/ChatGPT Work: generating a 400+ page custom textbook for a spring course and drafting a generative-AI handbook for a first-year business-school course. His core point is not that AI replaces instructional expertise, but that strong upfront context-setting, architecture design, and pedagogical review can compress the first-draft phase of course-material creation dramatically. The article also argues that productive-friction design matters inside AI-enabled materials: the handbook improved most when activities required students to try, predict, and critique before leaning on AI.
Pull quotes
Context is king
“The first step in this process was to provide Codex with the context it needed to do a decent job on the first pass.”
Productive friction strengthened the handbook
“The biggest improvement came when we revised the activities to take productive friction into consideration through some activities like having students generate their work before consulting AI, predicting what AI would come up with, critiquing AI output, and similar activities.”
Judgment remained the real work
“Although AI helped me draft and revise the handbook, much of the real work was judgment.”
Big ideas
- Learning still needs some struggle, even when AI can make things easier
- District AI work is a long-term redesign project
- AI is changing what knowledge work asks people to do
Claims
- Learning requires some productive struggle that AI can remove
- AI literacy should teach students what to do with AI, not just what to think about it
- District AI implementation needs living guidance and teacher-led redesign
- Agentic AI can preserve thinking when students have to design the work
Key evidence and examples
- Van Slyke gave Codex unusually rich context before drafting: a syllabus, a sample chapter from a successful co-authored textbook, a cognitive developmental hierarchy paper, diagrams, and chapter-level architecture.
- He spent substantial time up front on table of contents, pedagogical pattern, source and verification standards, intended audience, and chapter overviews before expecting good chapter drafts.
- After iteration on the first few chapters, he reports that later textbook chapters were often usable on the first pass, producing a 400+ page course-specific draft in about four hours of work with Codex.
- In the handbook experiment, the team built a five-skill architecture, audited the draft for voice evidence, student usability, and activity design, then strengthened it by adding productive-friction routines.
- The productive-friction revisions required students to do some work before AI help, predict likely AI responses, and critique AI outputs rather than passively accept them.
- Van Slyke explicitly argues that the durable value is not raw speed alone but the teacher judgment required to decide what students need, where AI could short-circuit learning, and how to design for responsible use.
Education relevance
Highly relevant for teacher professional development, curriculum design, district AI guidance, and assessment redesign because it offers a concrete pattern for using AI to accelerate bespoke materials while keeping pedagogical judgment and student cognitive work at the center.
Durability note
The named tool may change, but the durable lesson is that AI-supported curriculum creation improves when educators do substantial context engineering, specify instructional architecture, and redesign activities to preserve productive friction.
My notes
- Useful bridge source between faculty workflow change and K–12 policy/PD: it translates “context engineering” into a teacher-facing curriculum-design practice rather than a generic prompting trick.