How this was made

Where it started

In 2016 Jörn Dinkla published Die komplexe Perspektive (The Complex Perspective), a conventional book in German and English. It applied complexity science to economics, society, and politics. This site is its 2026 successor. It is not a new edition of the book. It is an interactive site, built for reading out of order and for running models yourself.

What the AI did

1. Research. Fourteen research briefs on eleven topics, all about what has changed since 2015, were written by Claude with web search. They cover agent-based modeling, economics, network science, data, the AI revolution, and digital politics. They total about 9,300 lines and are the main source material.
2. Drafting. Claude drafted every module from those briefs and from the 2016 book's chapters. It wrote in one voice, a defined persona called the analytical observer. The coloured asides in the modules are other personas: a realist on cost curves, an editor's plain-language restatement, an adversarial objection, and source notes.
3. Critique. Each module went through three AI critic loops. An adversarial editor attacked the argument. A comprehension editor checked whether a non-specialist could follow it. A method critic checked it against the 2016 book's way of explaining: earn each definition, build a small example first, and let the reader predict before the simulation runs. Each loop repeated until the critic stopped finding new problems.
4. Playtesting. AI agents role-playing different readers walked through the modules in order. They recorded where a concept was used before it was defined and where they would have given up.
5. Sourcing. A late pass by AI agents looked up the studies, reports and figures behind the claims, corrected the ones that did not match, and added a source note under each paragraph it could support. Every linked source was opened and checked; where a site could not be opened, the citation has no link.
6. Code. The site, the 30 interactive simulations, and their tests were written with Claude Code, Anthropic's coding agent.

What the author decided

The AI proposed. These choices were the author's:

• The subject, the four-part structure, and the order of the sixteen modules.
• Restoring foundations the AI drafts had dropped from the 2016 book, such as prices as information, reductionism versus emergence, and the data-science method. The new edition stays neutral on the book's original political argument.
• Splitting the original "human being" module into four, after the simulated readers named it the one place they hit a wall.
• Which structural recommendations from the critics and the playtests to adopt.
• English only, static hosting, no AI features on the site for now, and the licenses.

Known limits

• Sources were added late, and are not finished. A sourcing pass in October 2026, also run by AI agents with web access, added 259 citations and corrected about 150 claims that did not match their sources. About 60 claims could not be checked in that pass and are still open. A claim without a source note beside it may be one of them; treat it as a claim to check, not an established fact.
• The research is AI-written. The briefs were produced with web search, but they were not checked line by line. Errors in them may have carried into the modules.
• The simulations are toys. They are built for intuition, not prediction. None is calibrated against real data.
• The readers were simulated. The playtests used AI personas. Human reader feedback is still to come.
• It dates. The material reflects the state of the field in 2026.

Why say all this

Because the site's subject is the AI age, and it would be odd to hide how the site itself was made. AI authorship is stated openly here, as it is throughout the author's work. If you find an error, it is more likely to be a confident AI claim than a typo. Read the uncited parts with that in mind.

Start with the welcome module · See all modules