The physical edition. Every prior edition ran the loop on information; this one runs it on matter. Claude Code designed a wall frame for the bird station's 13.3" e-ink panel and its Raspberry Pi 5 — not by driving a CAD program, but by writing one: a 267-line build123d script whose dimensions were read off Pimoroni's official drawing, and which emits every part as STL and STEP, an assembly model…
The foundations edition. Every other page here tells you how to operate the tools; this one explains what is happening inside them, using the domain people find most magical. Sound is already numbers — 44,100 of them a second — and the whole breakthrough was compressing that into a few hundred learned symbols a second, short enough for a machine built to finish sentences to finish a song instead.…
The refresh edition. Two Anthropic models shipped since the last capture — Sonnet 5 on June 30, Opus 5 on July 24 — and one of them broke the rule this site's tier list was built on: for as long as these leaderboards have existed, the most capable model was also the priciest. Not any more. Opus 5 took the top of the independent index at LOWER cost per task than the model below it. Every board on…
The SEO reading. Most "agentic SEO" wins are self-reported traffic graphs that crater in six months. This chapter pulls the real receipts from two products built by AI agents end to end: Folderly EmailGen — a brand-new subdomain that hit #1 for "business email generator," 467K impressions, third-party traffic value climbing ~$0 → ~$320/mo — and LinguaLive, the cautionary tale whose 153K-impression…
The builder's eval, one altitude above Chapter 25. Chapter 25 was a three-line smoke detector for a skill's artifact; Chapter 47 is for when the model's output is the product itself. The Anthropic API course ships, as working code, the two tools most builders never build — a way to SCORE output and a way to RETRIEVE the right context. Both ported to TypeScript over a weekend: a PromptEvaluator…
The showcase. Chapter 46 argued that taste is the last mile; /good-taste makes it visceral — a side-by-side of generic-AI slop (purple glow, floating blobs, a fake dashboard) next to the same brief art-directed with a public skill. The skill is Leon Lin's open-source taste-skill (github.com/Leonxlnx/taste-skill, 45k+ stars), not mine; this page is what stealing a skill well, with credit, looks…
The design reading. Forty-five chapters taught you to build with AI; none taught you to make it look like something. Chapter 46 fills the gap with the book's own receipts — a contrast failure that sat in our light-theme tokens for weeks until the arithmetic caught what every eye had passed, flicked.email's three landing pages in three different typefaces, and a moon-base image set for Reach…
The fourth reading — the tier list gains an independent referee. Between the crowd's LMArena votes and the vendors' launch decks, Artificial Analysis runs its own agentic harness and prices every run, so you can finally sort the leaderboard by what a task actually costs, not just by who scores highest. Shipped: - New "What an independent referee measures" panel on /tier-list — Artificial…
The economics week — the research timeline gets the grid that turns "this plan can't be profitable on me" into numbers, and the tier list catches up to the Fable 5 era with a fresh LMArena snapshot and a new lab-claims panel. Shipped: - New research note (top of the timeline) — "The subscription subsidy, quantified." SemiAnalysis's Tokenomics model: a $20 claude-pro can draw ~$400/mo of…
The reference layer — five search-facing pages distilled from the corpus, each built on receipts the official docs can't print: the real bills, the configs that fired, the agents that broke. Plus the machine plumbing: every deploy now announces itself to the indexes ChatGPT and Claude actually read from. Shipped: - Five new reference pages, every claim traced to a chapter or a first-party source…
The Self-Audit — the multi-agent audit pointed at the layer nobody reviews: the agent's own configuration. Run the same evening Fable 5 landed, on the thesis that a capability jump is a standing audit trigger for the scaffolding under it. Shipped: - New method page: /self-audit — five parallel auditors (skill library, per-session context budget, memory system, hooks + permissions, automation…
Fable 5 / Mythos 5 — the withheld model shipped, and the Playbook files it the day it lands: a research note that corrects the May 6 call, a ten-page model file (incl. the system-card episodes), and the three evergreen pages the keyword map has been waiting for. Shipped: - New research note (top of the timeline) — "Fable 5 / Mythos 5 — the withheld model shipped, split in two." The May 6 note read…
Forty-five chapters. The App Store Without Swift — Claude Code one-shot a native iOS app for LinguaLive, and shipping it to the App Store meant walking through the gates the operator, not the agent, has to own. The proof is a RevenueCat renewal, read honestly. Shipped: - New Chapter 45 — "The App Store Without Swift" (Part V, The Building Site; Building Products section). The apex of the…
Learn — the on-ramp before the book. A new /learn page routes total beginners through the official free courses, in order, then hands them to the Playbook. Shipped: - New standalone page /learn — the entry point for anyone new to AI, Claude, Claude Code, or Cowork. A staged learning ladder (concepts → official courses → orient + set up → read and build), a five-word concepts primer that links the…
Forty-four chapters. Dreaming — the autonomous loop pointed inward: a local, propose-only twin of Anthropic's memory feature that surfaces candidate lessons from your own sessions and verifies every quote against the raw transcript, but never writes to memory itself. Shipped: - New Chapter 44 — "Dreaming — Memory That Curates Itself" (Part II, Working Memory; also in the General + Claude…
Dynamic Workflows — a standalone page on Opus 4.8's headline feature: the swarm that runs itself. Shipped: - New standalone page /dynamic-workflows — Opus 4.8 ships dynamic workflows: Claude writes a JS script that plans a big task, fans out up to 16 parallel subagents (1,000 per run), runs an adversarial verifier pass, and reports one answer. Part explainer (what it is, agent-teams vs workflows,…
Forty-three chapters. The Codex loop from Ch 42 runs for real — a 90,000-line product refactor, proof-checked. Shipped: - New Chapter 43 — "Codex as Saviour" (~2,390 words, 13-min read). The full-scale payoff of Ch 42's standing loop ("simplify, follow my design system"): pointed at a real shipping product (Folderly Email Generator), a bench of fourteen named explorer agents deleted a net 91,874…
Forty-two chapters. Codex gets the hands-on companion — second opinion, on a loop, proof-checked. Shipped: - New Chapter 42 — "Codex on a Loop" (~1,950 words, 11-min read). The hands-on companion to Ch 35's strategic day-shift/night-shift framing. The thesis: Codex isn't a better Claude Code, it's a second prior I run on a loop — pointed at Sentry (errors), PostHog (funnel-drop anomalies), and…
Every page one click from the homepage. Tile grid grows to 3 rows. Shipped: - Homepage tile grid expanded from 8 → 12 tiles (3 rows of 4). Row 3 added: Swarms · GitHub Playbook · Setup showcase (62 skills · 32 agents · 12 plugins) · Research notes (what shifted this week). The pages I shipped this week are now first-class tiles, not secondary-row text links. - Hanging-pages cleanup — secondary…
Forty-one chapters. The HTML-ization thesis gets its own chapter slot. Shipped: - New Chapter 41 — "Send the Link, Not the File" (~1,800 words, 9-min read). Chapter-length narrative companion to /html-first. The story version of the HTML-ization thesis: every report, pitch, audit, deck, model in the portfolio ships as a live interactive link in a private repo, not a PDF or slide attachment. Same…
Swarms — the deep version, with the skills and the prompts and the receipts. Shipped: - New /swarms page (~3,800 words) — the operator's deep dive into multi-agent orchestration. Embeds the SwarmVisualizer widget (4 patterns animated: fan-out, pipeline, map-reduce, adversarial). The 5 × 4 wave pattern explained with the real five-wave breakdown from /swarm-strategic-plan. - The skill shelf — 10…
Prompting is basic now. The leverage moved up the ladder. Shipped: - New Chapter 40 — "Prompting, or the Knob You Probably Shouldn't Tune" (~3,200 words, 12-min read). The contrarian take: prompting is a basic skill now, the real ladder is skills → swarms → memory → data layer. Opens with the verbatim "rigor enforcer" prompt (works, but the wrong leverage) and contrasts it side-by-side with…
HTML-ization, expanded — the recipe, the gallery, the receipts, the edges. Shipped: - /html-first expanded from 4 sections to 9: kept the existing case (AFC + Folderly audit + Do-this-Monday) and added five new ones - New section "More across the portfolio" — the order of operations (pitches first, audits next, internal deliverables third, communication fourth), the one-to-two gap that flips the…
The launch as its own experiment in the thesis. Shipped: - New /launch-week page — the launch is its own forwardable artifact. Numbers panel (live GitHub stars, newsletter open rate, subs delta, page-views), a day-by-day timeline of every distribution surface, a "what I'd do differently" section that populates from real data not Day-0 guesses, a meta-callout explaining why a PDF version would be…
Two stacks, not one. The sovereign tier is built. Shipped: - New /sovereign-stack page — the open-weights answer to the deprecation calendar. The Onyx Open-Source LLM Leaderboard rebuilt as a tier list (Overall / Coding / Math / Chat / Reasoning), Ollama vs LM Studio vs vLLM, five hardware tiers ($2k MacBook → $40k+ deploy) with what-breaks-first per tier, quantization + KV-cache math,…
The book is open. The source is the recipe. Shipped: - Repo flipped to public after a full git-filter-repo history scrub — 71 commits rewritten, 0 leaks across blobs and commit messages from an independent fresh-clone verification, backup ref preserved server-side - New /launch page — the launch artifact IS the demo: typewriter hero, animated stat odometer…
The method, embedded. Click the artifacts. Shipped: - New /html-first page — the htmlization thesis (every deliverable ships as a live interactive artifact, not a dead file) with two REAL artifacts embedded and clickable - AFC case: the dinner-table idea whose investment deck + robot-stable annex got spun up as interactive HTML before the next meeting — Vlad's own venture, embedded as-is -…
The bench moved. The reports stopped being files. Shipped: - Gemini 3.5 Flash + Claude-for-the-legal-industry logged as dated research notes — signal-vs-receipt discipline, not a leaderboard edit - Ch 35: corrected the stale "Gemini 3 Pro" reference + added the Flash-beats-last-gen-Pro signal, scoped as a Ch 29 cost question rather than a re-tiering - Ch 29: new section — "The price of a model is…
Retitled for navigation. Voice survived. Shipped: - All 36 chapter titles rewritten — topical-primary, with question form on 9 chapters where intent shape beats subject shape - Old hype titles preserved as subtitles ("The Day I Killed My Tabs" now sits below "AI as an Operating System") - Cmd-K resolves "cron" → Scheduled Tasks, "browser" → Browser Agents with Playwright, "permissions" → When to…
36 islands became one six-act journey. Shipped: - /how-to-read — ~1,100-word prologue. Who this is for, the reframe, the journey, three ways to use the book. - /journey — 6-part narrative arc (Reframe → Memory → Workshop → Discipline → Building → Frontier) - /questions — 18 questions Vlad answers most weeks, each with short answer + chapter pointers - Part pill on every chapter hero — readers know…
Closed the original brief. Six new chapters, two new pages. Shipped: - Ch 31 — The Stages: Ideation → Foundation → Creation → Polishing → Security → Deploy - Ch 32 — How to Build Rick (OpenClaw, NemoClaw, Hermes archetypes) - Ch 33 — Browser Agents with Playwright (login, click, scrape, post) - Ch 34 — Persona Agents and the Four NEVERs - Ch 35 — Codex × Claude Code (day shift, night shift) - Ch…