Paperclip im Maschinenraum: Agenten arbeiten, Lena prüft

21. Juli 2026

Part of Skaldify is not created in the visible shop, but in the internal engine room. Paperclip is being tested there as a task and agent system: for literature work, product ideas, technical checks, graphic briefings, POD concepts and structured drafts.

The conclusion is deliberately sober: agents can start a lot of work, but they do not replace review. Good results appear where tasks are narrowly framed, context is limited and Lena curates the output.

What Paperclip can do well

It can generate idea lists and variants, deliver first raw drafts, suggest structures and scene anchors, prepare technical tasks in parallel and collect material that can later be checked.

Larger work waves also showed a technical lesson: long model answers need a stable bridge. A non-streaming connection ran into timeouts for longer answers. Switching to streaming made the process more robust.

The most important rule

Paperclip may work, but Skaldify does not publish unchecked agent output. Raw material is valuable when it is treated as raw material. The public result must be reviewed, selected, edited and documented.