# Recommended videos

One opener per day and one video per lesson, embedded on the lesson pages. None is required to complete an exercise.

**How this list was made.** Candidates came from research notes and YouTube search results; each pick was chosen by recency first and reputation second, from original or conference channels only. Every link was verified mechanically (the video exists, is embeddable, is not a live stream, has an upload date) and its chapters were confirmed against the description on 2026-09-22. `build/checks/check-videos.sh --manifest build/state/videos.json` re-checks them.

## Day 1: From chat to spec

| Slot | Title | Channel, date, length | Watch | Notes |
| --- | --- | --- | --- | --- |
| Opener | Andrej Karpathy: From Vibe Coding to Agentic Engineering w/ Stephanie Zhan | Sequoia Capital, 2026-04-29, 29:49 | [15:46–25:17](https://www.youtube.com/watch?v=96jN2OCOfLs&t=946s) | Chapter 'From Vibe Coding to Agent Engineering' (15:46 to 25:17): what changes when the model gets tools and a loop, and why the engineer's job becomes directing and verifying rather than typing. |
| Chat vs agent, and the four-part prompt | How Claude Code Works | Claude, 2026-05-13, 2:50 | [whole video](https://www.youtube.com/watch?v=6bs5b4FltCU&t=0s) | Whole video (170s): the agentic loop that gathers context, takes action and verifies the result, plus the context window, tools and permission modes that separate an agent session from a chat session. |
| Chat vs agent, and the four-part prompt (also watch) | Boris Cherny: We Cut 80% of Claude Code’s Prompt | Y Combinator, 2026-07-27, 35:51 | [19:32–24:42](https://www.youtube.com/watch?v=qyPCVqFUyDo&t=1172s) | Chapters 'Prompt Engineering Is Changing' (19:32) and 'The Two-Week Claude Code Prompt' (21:57): how the shape of a good prompt changed, and the long-lived prompt they iterated on for two weeks — goal, bounds and verification rather than phrasing tricks. (Shares the video with lesson 6, different chapters.) |
| Seeing the diff before the diff with plan mode | The Explore → Plan → Code → Commit workflow in Claude Code | Claude, 2026-05-17, 3:11 | [whole video](https://www.youtube.com/watch?v=xJQuF02NAK8&t=0s) | Whole video (191s): the Explore → Plan → Code → Commit workflow — research before coding, define success criteria, then implement, with subagent review before pushing. |
| The spec and its criteria | Why Software Factories Fail | AI Engineer, 2026-07-23, 19:17 | [14:58–19:17](https://www.youtube.com/watch?v=Ib5GBkD555M&t=898s) | 14:58 to 19:17 (259s), chapters 'Turning the lights back on: plan up front' and 'Too many bad PRs, and closing advice': Dex Horthy on what he does instead of the lights-off factory — product review, system architecture, program design down to types and call graphs, then vertical slices, with the claim that thirty minutes of alignment up front saves hours of review. |
| The spec and its criteria (also watch) | Spec-Driven Dev Is Back. But Not How You Think • Daniel Terhorst-North & Gojko Adzic • GOTO 2026 | GOTO Conferences, 2026-04-27, 39:03 | [10:39–14:39](https://www.youtube.com/watch?v=6mLYZF97oaU&t=639s) | Chapter "We've seen this before" continued, 10:39 to 14:39: why acceptance criteria that cannot fail are the part of a spec that survives; four minutes, outside the lesson's sixteen. |
| Context is a budget | Boris Cherny: We Cut 80% of Claude Code’s Prompt | Y Combinator, 2026-07-27, 35:51 | [3:21–10:30](https://www.youtube.com/watch?v=qyPCVqFUyDo&t=201s) | Chapters 'Why Claude Code Deleted 80% of Its System Prompt', 'Press Delete on Your AI Product', and 'How to Rebuild Your System Prompt' (03:21 to 10:30): the ablation argument for shorter instructions, which is the lesson's case for a short CLAUDE.md. |

## Day 2: Loops and your environment

| Slot | Title | Channel, date, length | Watch | Notes |
| --- | --- | --- | --- | --- |
| Opener | Claude Fable, Claude Tag, and Anthropic's Culture — Cat Wu & Thariq Shihipar ft Simon Willison | AI Engineer, 2026-07-15, 51:30 | [30:57–37:53](https://www.youtube.com/watch?v=uU5Gv2h8-9g&t=1857s) | Chapter 'Safety, security, and using Auto Mode' (30:57-37:53): Cat Wu and Thariq Shihipar on Auto Mode as the recommended way to run long-running agentic tasks, and the evals and red-teaming (prompt injection, data exfiltration) that make running an agent unattended safe. |
| Spec, plan, execute, verify, and the four places drift gets in | Learnings from a No-Code Library: Keeping the Spec Driven Development Triangle in Sync | Drew Breunig, 2026-03-04, 40:01 | [17:35–22:34](https://www.youtube.com/watch?v=8TXAlOFkmk0&t=1055s) | 17:35 to 22:34, the chapter 'The spec-driven triangle': spec, tests and code as three corners that drift apart unless each is checked against the others. |
| The runner: non-stop until a stop condition, with limited gates | Why Software Factories Fail | AI Engineer, 2026-07-23, 19:17 | [5:52–8:56](https://www.youtube.com/watch?v=Ib5GBkD555M&t=352s) | 5:52 to 8:56 (184s), chapters 'The agentic factory and turning the lights off' and 'Why it fails: the July 2025 lights-off experiment': Dex Horthy on the unattended loop with no human reading code, and why it fell apart. |
| Recursive self-improvement: the output, the instructions, the workflow | Why The Harness Matters More Than The Model / YC Paper Club | Y Combinator, 2026-09-07, 60:11 | [13:56–18:35](https://www.youtube.com/watch?v=n9xKblqyQ28&t=836s) | Chapter 'Self-improving harnesses' (13:56-17:22, running to the speaker handoff at 18:35): the case that the harness, not the model, is what gets improved, and how a harness that rewrites its own scaffolding compounds. |
| Your environment: profile, permissions, write path | Beyond the basics with Claude Code | Claude, 2026-05-22, 47:19 | [0:00–5:00](https://www.youtube.com/watch?v=tuY2ChJIx48&t=0s) | The first five minutes of Anthropic's own session on CLAUDE.md, MCP, skills and auto mode: the settings and permission surface that defines what the agent is allowed to do. |
| Your environment: profile, permissions, write path (also watch) | #17 - How Zalando Engineers Software & AI at Scale - Bartosz Ocytko (Executive Principal Engineer) | bynd code, 2026-09-20, 93:43 | [35:19–40:19](https://www.youtube.com/watch?v=1wIHIwS2EWo&t=2119s) | Chapter 'Enabling AI Engineering: Models, Tools & Agent Platforms' (from 35:19): Zalando's internal agent platform - the internal LLM platform and MCP servers engineers get, and how access is handed to agents without each team wiring its own credentials. |

## Day 3: Choose your adventure

| Slot | Title | Channel, date, length | Watch | Notes |
| --- | --- | --- | --- | --- |
| Opener | #17 - How Zalando Engineers Software & AI at Scale - Bartosz Ocytko (Executive Principal Engineer) | bynd code, 2026-09-20, 93:43 | [0:00–10:00](https://www.youtube.com/watch?v=1wIHIwS2EWo&t=0s) | Chapter 'How Zalando Scales AI Engineering' (00:00-14:22), trimmed to the first 10 minutes: how one engineering organization moved to agentic coding, what it standardized and what it gave up. |
| Why a framework, and what they share | #17 - How Zalando Engineers Software & AI at Scale - Bartosz Ocytko (Executive Principal Engineer) | bynd code, 2026-09-20, 93:43 | [35:19–40:19](https://www.youtube.com/watch?v=1wIHIwS2EWo&t=2119s) | Reuses the opener, different segment. Chapter 'Enabling AI Engineering: Models, Tools & Agent Platforms' (35:19-50:46), first 5 minutes: the shared access paths and centrally distributed agent tooling the organization standardized on. |
| Why a framework, and what they share (also watch, optional) | Spec-Driven Dev Is Back. But Not How You Think • Daniel Terhorst-North & Gojko Adzic • GOTO 2026 | GOTO Conferences, 2026-04-27, 39:03 | [6:34–10:39](https://www.youtube.com/watch?v=6mLYZF97oaU&t=394s) | Chapter "We've seen this before (And it didn't end well)" (06:34-10:39): Terhorst-North and Adzic place today's spec frameworks against CASE tools and model-driven architecture, separating what every spec framework claims from what survives. |
| Meet the three: OpenSpec | OpenSpec GitHub Tutorial: Spec-Driven Development for AI Coding Assistants | Alex Hitt, 2026-07-01, 8:48 | [1:11–6:04](https://www.youtube.com/watch?v=mNlXDfAa22k&t=71s) | 1:11 to 6:04 of the 8:48 walkthrough: brownfield delta specs and the changes-versus-specs layout, before the command tour. |
| Meet the three: LID | (no video) | | | No walkthrough of Linked-Intent Development was found. The LID tab teaches from the project's README and your own mapping run in Exercise 1. |
| Meet the three: BMAD | BMad V6 is Finally Here… /bmad-help, /party-mode... Pure Magic | BMad Code, 2026-02-23, 24:52 | [6:34–11:23](https://www.youtube.com/watch?v=4VPoGSeI2sw&t=394s) | 6:34 to 11:23: the installer walkthrough (npx bmad-method install, choosing modules and tools). |
| Pick by the failure you actually have | Simon Martinelli - Lessons from Spec-driven Development - AI Native DevCon June 2026 | AI Native Dev, 2026-07-22, 31:38 | [0:00–5:00](https://www.youtube.com/watch?v=odbNXv9xXjc&t=0s) | The first five minutes of the talk: the failure that started him on specs (a system built fast with an AI tool that he could neither explain nor adapt when a second customer arrived), then the tools he judged too developer-centric and the process he built instead. |
| Adopting without a rewrite | OpenSpec on a Real Project: What the Demos Don’t Show | Dmitrii Balakin, 2026-07-18, 11:32 | [7:46–11:32](https://www.youtube.com/watch?v=YHPMs252Opc&t=466s) | Chapters 'What OpenSpec Gets Right' (7:46) through 'Where I Landed' (11:09): what review-before-code bought him, the cost of too many Markdown files, where specs drifted from code, and where he landed. |