How we use AI

Who does what

Diego Garrido decides what the site covers, approves each story’s angle, judges whether its evidence holds, approves it for publication, and is responsible for everything it publishes.

AI agents (Anthropic’s Claude models) do most of the work under his direction: they search for and read sources, build the evidence for each story, draft text, check facts, review drafts, make graphics, and build pages. Different agents take different roles, and the agent that checks a story works independently of the agent that wrote it.

Safeguards

• Nothing an AI system remembers counts as a source. Every fact comes from a document an agent opened and a reader can open too.
• Every quotation is matched word for word against its archived source, and every number is recomputed.
• A separate checker re-finds sources from scratch, without seeing the writer’s reasoning.
• The agent that writes a story cannot add facts that are not in its checked evidence.
• No story is published without the human editor’s approval.
• No generated image is ever presented as a photograph, document, or screenshot.

Each story says how it was made

Every dispatch carries a short note on how it was reported: its main sources, its methods, and which tasks AI systems performed.

Conflicts

The agents run on Anthropic’s models, and Anthropic is a frequent source and subject of this site’s reporting. Stories about Anthropic are held to the same standard as any other, and say so.

Reflexivity

The program holds that descriptions of agents enter their environment and their training, so findings must survive their own publication. This site is such a description, written in part by agents of the kind it describes, working under subsidy. Its findings have to remain true after agents have read them.