Keyboard shortcuts

Press or to navigate between chapters

Press S or / to search in the book

Press ? to show this help

Press Esc to hide this help

Learning Tracks

Each track links roughly ten entries in a suggested order, chosen for a particular kind of reader. They’re highlights, not homework: start wherever your situation fits, follow the path as far as it serves you, and leave the moment a link points at what you actually need.

Your First Day with an AI Agent

For anyone who has never directed an AI coding agent. This path builds the mental model from the ground up: what a model actually is, how to prompt it, and why the context window shapes every choice. From there it makes the jump to agents that act through tools, sets up an instruction file so the agent remembers your project, and closes on the two habits that keep you in control: the verification loop, and deciding how much human supervision the work really needs.

  1. Model
  2. Prompt
  3. Context Window
  4. Agent
  5. Tool
  6. Instruction File
  7. Verification Loop
  8. Human in the Loop

Building Things That Work

For readers who want the software-construction foundations that hold a system together — the vocabulary experienced developers reach for without thinking. It names the problem, turns it into requirements, then works down through architecture and its pieces: components, the interfaces and boundaries between them, and the cohesion and coupling that tell you whether a split is sound. The path lands on the principles that organize all of it, abstraction, separation of concerns, and decomposition, and on the test that proves the result actually works.

  1. Problem
  2. Requirement
  3. Architecture
  4. Component
  5. Interface
  6. Boundary
  7. Cohesion
  8. Coupling
  9. Abstraction
  10. Separation of Concerns
  11. Decomposition
  12. Test

Keeping Software Honest

For intermediate readers focused on making software correct and keeping it correct as it changes, with security treated as part of the same job. It starts with what "correct" even means: invariants, the tests that check them, the oracle a test judges against, the harness that runs tests at scale, and the regressions that keep a fix from quietly coming undone. Then it widens from accidental bugs to deliberate threats, walking the security line from threat model and trust boundary through input validation, least privilege, and the sandbox that contains whatever still gets through.

  1. Invariant
  2. Test
  3. Test Oracle
  4. Harness
  5. Regression
  6. Threat Model
  7. Trust Boundary
  8. Input Validation
  9. Least Privilege
  10. Sandbox

Mastering the Agentic Workflow

For readers who already have the basics and want to direct agents well on real projects. It leads with the highest-leverage skill, context engineering, plus the compaction that keeps long sessions from degrading, then scales the work out across a thread per task, subagents, and the parallelization that runs them at once. From there it builds the operator's toolkit of plan mode, skills, hooks, memory, and worktree isolation. It closes on the governance that keeps autonomy honest: an approval policy for what needs sign-off, and evals to tell you whether any of the tuning actually helped.

  1. Context Engineering
  2. Compaction
  3. Thread-per-Task
  4. Subagent
  5. Parallelization
  6. Plan Mode
  7. Skill
  8. Hook
  9. Memory
  10. Worktree Isolation
  11. Approval Policy
  12. Eval

From Idea to Product

A cross-cutting path that follows a raw idea all the way to live software, drawing from several sections at once. It begins where every product should, a problem worth solving, then sharpens the product thinking: who the customer is, why they would choose you, and the user stories and acceptance criteria that turn a need into work an agent can do. From there it shifts to shipping: deployment, continuous integration, feature flags, and the rollback that undoes a bad release fast. It ends on the observability that tells you whether the thing you shipped is actually working.

  1. Problem
  2. Customer
  3. Value Proposition
  4. User Story
  5. Acceptance Criteria
  6. Deployment
  7. Continuous Integration
  8. Feature Flag
  9. Rollback
  10. Observability