60 The Linear Plan
The Linear Plan
This page is the editorial working document for the book. It explains the audience, the structure, the chapter shape, and the open follow-ups. Readers can skip it; authors and contributors should not.
60.1 Audience and assumed background
The target reader is a second- or third-year undergraduate in economics, finance, business, analytics, or PPE — or a first-year MA/MSc in the same. They have:
- One solid course on data analysis, ideally Békés–Kézdi (2021). The textbook is treated as a parallel companion, not a prerequisite the book re-derives. Where a chapter brushes a method (DiD, IV, fixed effects, joins, panel structure), it links back to the relevant Békés–Kézdi chapter rather than re-explaining.
- A working knowledge of Python — enough to write a small script and read someone else’s.
- A chat-AI account they have used casually but not yet for analytical work.
We do not assume Git, GitHub, terminal fluency, an IDE, or API experience. Part 0 covers all of that.
60.2 What this book is — and isn’t
This is a standalone book, not a companion website. Each chapter is meant to be read on its own, on the page, without clicking elsewhere. The companion course at gabors-data-analysis.com/ai-course/ remains live for instructors who want the weekly-class form of the same material; the book does not depend on it.
Editorial consequences:
- No pictures by default. A diagram only earns its place when the prose cannot. Screenshots date in months and are avoided.
- Layout simplified. The course pages used hero blocks, “week cards”, and other web-native ornaments. The book strips these. One simple chapter shape.
- Self-contained chapters. A reader who opens the book at Chapter 14 should be able to read it without first reading Chapter 13.
- External links sparingly. When we link, we link to durable references (Békés–Kézdi sections, Anthropic / OpenAI documentation, foundational papers), not to course pages.
60.3 The chapter shape
Every chapter follows the same skeleton, with two flavours (Learn or Practice) and three length budgets (1–2, 2–4, or 4–5 pages). The full specification lives in chapter-template.qmd.
The non-negotiable parts of every chapter:
- Opening hook — two or three sentences, no throat-clearing.
- What you will get from this chapter — three to five concrete outcomes.
- Body — concrete examples first, generalisations second.
- Where AI helps · Where AI bluffs — specific to the topic, never generic.
- Keep this with you, not the AI — the decisions that stay with the human.
- Try this — required for standard and long chapters; optional for short.
- AI and me — the same three reflection questions every chapter.
Voice: precise but plain, short sentences, no bluffing, engaging without being breathless. Examples drawn from health and food, sports (especially football), global politics, finance, international business.
60.4 The eight parts (plus front and back matter)
| Part | Focus |
|---|---|
| Front matter | Preface, How to use this book, Edition + model snapshot |
| Part 0 | Tools and harnesses — git, GitHub, Python, AI account, plus a “three harnesses” overview, terminal, VS Code+Copilot, and CLI agent |
| Part I | What an LLM is, and what it isn’t |
| Part II | Working with AI: data foundations — documenting, joining, reporting, graphing |
| Part III | Working at scale: agents and pipelines |
| Part IV | Text as data |
| Part V | AI in empirical research — controls, instrumental variables |
| Part VI | APIs and automation |
| Part VII | Capstone — Manager Impact in Football |
| Part VIII | Labs — work on your own |
| Part IX | Working honestly — academic integrity, cost and budget |
| Reference | Case studies, Where to go from here, this Linear Plan |
| Appendix | Rights, acknowledgements, thanks |
Harness-agnostic by design. Earlier drafts pushed the CLI to Part III. The current structure introduces all three harnesses — chat, IDE+Copilot, CLI agent — in Part 0, so the reader can pick whichever one they want and use it across the whole book. Chapters from Part II onwards are written so the workflow runs in any of the three; where the harness materially changes the answer, the chapter says so.
The current draft is a mix of newly written chapters (drafted to the chapter template) and inlined source content from the live course (slated for rewrite). Placeholders sit in slots where the book argues a chapter is needed but the source course did not provide one.
60.5 Migration status
The book was first scaffolded with {{< include >}} shortcodes pointing at the course’s source files. After the move to this standalone repository, those includes were inlined: every chapter file is now self-contained.
Drafted to the chapter template (Spring 2026 edition, first wave):
- Part 0:
00a-prep,00b-git,00c-github,00d-python,00e-ai-account,00f-three-harnesses,11-terminal-basics,13-vscode-copilot,12-install-cli. - Part I:
01-ai-for-coding,02-llm-review,03-which-ai,04a-prompting-basics,04b-where-ai-fails. - Part II:
05-documenting,07-joining-tables. - Part IX:
89-academic-integrity,89a-cost-and-budget.
That is 18 chapters of real prose, totalling roughly 25,000 words and 35–50 book pages.
Awaiting rewrite (currently inlined course content; structure stands but voice needs the chapter-template pass):
- Part I:
04-glossary(already a useful reference; light touch only). - Part II:
06-doc-fundamentals,08-reports,09-ideas-good-report,10-creating-graphs. - Part III:
14-agentic-cli,15-advanced-cli,16-data-to-report,17-designing-projects,18-reproducible. - Part IV:
19-nlp-basics,20-text-analysis,21-sentiment,22-pdf-guide. - Part V:
23-ai-research-controls,24-ai-research-iv. - Part VI:
25-api-keys,26-api-intro,27-llm-apis-python,28-api-advanced,29-walkthrough-wb-fred,30-walkthrough-fbref. - Part VII:
31-capstone-brief,32-capstone-data,33-capstone-text,34-capstone-did. - Reference:
90-case-studies,91-beyond. - Appendix:
A1-rights-thanks.
Open follow-up tasks, in rough priority order:
- Per-chapter rewrite for the inlined content. Apply chapter-template.qmd to each. Strip web-only artefacts (hero blocks, week-card divs, image references). Add the Helps · Bluffs and Keep this with you sections.
- Internal-link audit. Inlined content still references course-website paths (
/da-knowledge/...,../week08/index.qmd). Rewrite to chapter anchors inside the book. - Strip remaining R snippets. Python only.
- Drop the
images/folder when the per-chapter rewrites no longer reference it. - Image-free check. A final pass to confirm no
references survive.
60.6 Editorial decisions (locked for Spring 2026)
- Audience. Late-undergrad / first-year postgrad in economics, business, finance, analytics, PPE.
- Companion textbook. Békés–Kézdi (2021). The book points to it for methods rather than re-deriving them.
- Form. Standalone book. No companion website dependency. Web-only delivery (HTML), but readable as a single document.
- Pictures. No pictures by default. Diagrams only when they explain something prose cannot.
- Code language. Python only.
- CLI agent. Claude Code is the working tool; alternatives get one-paragraph mentions.
- Versioning. Yearly spring edition, dated and labelled. Inter-edition errata in the GitHub issue tracker.
- Chapter length. Variable. 1–2 pages for a single idea; 2–4 for a workflow; 4–5 sparingly for an end-to-end walkthrough. No artificial uniformity.
- Chapter structure. One template, two flavours (Learn / Practice). Same skeleton; different weights.