Agents - Long Story Short

This research introduces a new approach to automated journalism using AI agents to personalize news production pipelines. Grounded in the principles of modular journalism and structured content, the system delivers tailored information based on each user's specific needs and interests.

The pipeline consists of five autonomous units, each staffed by AI agents with a specialized role. It begins with a structured news health check that assesses content usability and identifies core informational value. The next unit detects rhetorical distortion and editorial effects. A third unit maintains the inventarium – a structured record of verified information – while a reporter agent identifies missing elements and escalates micro-assignments to human journalists when needed.

The final unit, the cautious generator, assembles validated modules into complete stories under strict constraints. Its core principle is clear: don't fabricate. Ambiguities are preserved. Absences are acknowledged, not filled.

The system follows a human-in-the-loop model, with editors performing final review and stepping in only when agents encounter issues they can't resolve. This collaborative architecture ensures high-quality outcomes and builds trust in AI as a tool – not a substitute – for journalistic judgment.

The goal is to move beyond static story formats and toward a responsive, adaptive system – a learning organism – continuously refined through human feedback and accumulated editorial precedent.