# Autopilot, Not Copilot

> Research, curation, planning, reporting, drafting, and layout run as one pipeline—with human proofing before anything ships.

By Papyrus Editorial Team · Published 2026-05-22

Canonical: https://read.p.apyr.us/papyrus/papyrus-agent-workflow/

A copilot waits for the next command. An autopilot follows a course that humans have set, adjusting within known boundaries and returning control when judgment is required. Papyrus is built on the second model.

Research agents gather source material and produce candidate evidence. Human curation decides what enters the trusted graph and what stays out of scope.

Editor agents use that graph plus policy to prioritize coverage, spot movement in signals, and assign reporting where the newsroom should invest attention.

Reporter agents collect context and structured findings. Copywriting agents turn reporting into draft stories aligned with style and policy. Layout agents package approved copy into edition pages.

Human review remains the final gate before publication. People can inspect references, redirect assignments, tighten policy, reject weak drafts, and approve only after the newsroom has shown its work.

Because the output is an edition—not a feed—each stage must choose. What belongs in this issue? What can wait? What context does a reader need before moving on?

The pipeline is deliberately specialized. Research agents do not publish directly. Editor agents do not pretend to be reporters. Layout agents do not invent coverage priorities. Each handoff passes structured context to the next.

The result is repeatable production with traceability from source curation to final approval—and automation that leaves room for interviews, field reporting, and editorial calls only humans should make.