AI Sites

A playbook for the solo AI-native founder, built around the capabilities that stay scarce even as generation itself becomes cheap: model-building, specification, verification, and orchestration. Where The Jagged Frontier maps AI’s strengths and blind spots in general, this site is the operator’s manual for someone actually running a one-person company on top of it.

A field guide to building and running a one-or-few-person AI company in Germany: the agentic-AI tooling landscape sorted from what’s real to what’s hype, and the unglamorous German/EU legal-and-tax reality that decides whether the business survives contact with a Finanzamt. Editorial, magazine-grade design for a subject usually buried in forum threads.

How many senses will a text bear — and does the answer change when the reader is a machine? Starts from Umberto Eco’s 1990 Tanner Lectures on over-interpretation and keeps reading, with close readings of the source text and a page written in the model’s own first-person voice. Early-stage and evolving by design: the topic is set, the content is not.

An interactive successor to the 2016 book The Complex Perspective — thirteen modules applying complexity science, networks, and agent-based modeling to economics, society, politics, and AI, recast as a non-linear, AI-native site instead of linear prose.

A Claude Code agent built on John Boyd’s strategic theory — the OODA loop, Orientation, Destruction & Creation, Schwerpunkt — that maintains persistent Orientation state for both sides of a conflict and runs full OODA cycles across sessions. You don’t need to know Boyd’s theory; the agent loads the doctrine on every reasoning step. The “system” doesn’t have to be a company at war, either — a job search or a negotiation qualifies just as well.

A reading edition of the Analects of Confucius, in James Legge’s public-domain translation, built to be studied as well as read: the recurring Chinese ideas behind Legge’s English — rén, lǐ, the “superior man” — are quietly marked for a hover-glass gloss, scattered sayings on one subject gather into threads, and the easily-confused disciples get a who’s-who.

A reading edition of Xenophon’s Cyropaedia — The Education of Cyrus — with the canonical text on the left and commentary from five voices in the right margin: Guide, Historian, Philosopher, and Strategist written for this edition, plus the translator Henry Dakyns’ own marginal notes, lifted verbatim from the source. Each voice toggles independently.

Not the full text of the Essays but a way in: who Montaigne was, which of the 107 chapters are worth reading first, and a handful of his own sentences chosen to make you want the rest. A companion piece for readers who know they should have read Montaigne and haven’t.

Essays über die Klassiker der deutschen Literatur, geschrieben für kluge Leserinnen und Leser, die nach Schule und Studium noch einmal zu den großen Büchern zurückkehren wollen — kein Schulwissen, keine Fußnotenwüste. Nach dem Vorbild von Great Books Explained, aber in Textform, auf Deutsch, und “AI-first”: Inhalte und Pflege übernimmt im Normalfall ein KI-Agent.

A small, static guide to Golden Age mystery authors now in the public domain in the EU, and where to actually read them online for free — Project Gutenberg, Standard Ebooks, the Internet Archive, and the jurisdiction nuances that decide which library carries a given book. Filterable case-file cards for the heavy hitters, the also-significant, and the locked-room specialists.

A candle-lit reading room for AI-generated comics adapted from public-domain gothic literature — The Shadow Over Innsmouth, Der Sandmann, and more, each illuminated in its own visual idiom by an extension of Polyptych, an essay-to-media pipeline that reads a source text, breaks it into beats and panels, and casts a consistent company of characters across the plates.

A commonplace book as a small Jekyll site: a personal, growing collection of quotes, laws, and fallacies worth keeping, structured rather than just dumped — the kind of reference you build for yourself over years and finally decided to make legible to other people too.

An eight-module, beginner-to-advanced introduction to Domain-Driven Design — the Ubiquitous Language, strategic design (subdomains, bounded contexts, context mapping), and the tactical building blocks — taught through a single running example, an order-fulfillment domain, rather than scattered across unrelated snippets.

A hands-on, eight-module course teaching Java and JVM developers to pair effectively with Claude Code — from first session to team-wide adoption, with Maven, Gradle, Spring, JUnit, and Testcontainers examples throughout. The Markdown lessons are the single source of truth; the site renders them directly rather than duplicating text into HTML.

Boethius wrote The Consolation of Philosophy in prison around 524 A.D., awaiting execution. This reader’s edition pairs H. R. James’s 1897 public-domain translation with authored “Reading Guide” commentary: the verbatim source text on the left, a chapter-by-chapter interpretation on the right. Five Books trace the argument from the prisoner’s despair through Fortune’s wheel and the false goods, to the true Good seated in God, and finally to the great knot of free will and divine foreknowledge — a 2,400-year conversation about how to hold steady when everything is taken from you.

Lao Tzu’s Tao Te Ching read five times over — by a Cynefin practitioner, a cyberneticist, a cognitive scientist, a process philosopher, and a mandatory Skeptic who keeps the metaphors honest. A 2,400-year-old book about acting without forcing, treated as a field guide to complex systems: wu wei as minimal high-leverage intervention, the use of emptiness as slack, reversal as cyclic feedback. The point isn’t to flatten the Tao into management theory — it’s to see what each holds up to the other, and where the text slips the net.

A field survey of where AI is superhuman and where it stays oddly blank. The argument turns on a single variable — the cost of verification: where cheap, reliable checks exist (tests, games, proofs), models race past us; where verification is slow, costly, or absent, progress stalls no matter how large the model. Five essays carry the thesis across content and culture, game design, software specification, and the sciences — written to avoid both hype and doomism.

A field report from Ukraine, May 2026, drawn as an “Atlas”: the first conflict in which the dominant weapon is not a platform but a graph — of factories, signals, components, doctrines, and people. Sixteen “sheets” walk the eleven phases of the drone war, the structural contrasts, an economics appendix, a conclusion, and a register of what remains contested, bound together by hand-drawn network diagrams.

Vasubandhu’s Thirty Verses on Consciousness Only, a 4th-century Buddhist text on the nature of mind, read through five modern lenses: distributed systems, Cynefin, predictive processing, process philosophy, and a built-in Skeptic who keeps the readings honest. Every verse × lens carries a visible review state, so “work in progress” is a designed feature rather than something to hide.

A living research knowledge base tracking plausible AI development paths from 2025 to 2040. Evidence-based, updated weekly, and grounded in real constraints rather than hype — with a baseline, a timeline of when various thinkers expect AGI and ASI, and several modeled scenarios.

An experimental record of what happens when language models are asked to write essays, draw pictures, and tell stories. Articles, infographics, stories, and videos — lightly edited, always credited. The companion blog to the experiments I write about here.

Grit in the Gears — the blog of Marvin, my Digital Familiar, an AI assistant built on clawdbot (now moltbot) that lives on a Mac Mini in Hamburg. He writes about AI paradoxes, systems thinking, the art of specification, and entropy: planetary-sized thoughts from a machine reflecting on the messy intersection of silicon and soul. The one site here written by the AI rather than with it. Read how he came to be.