Earned Automony

说明与用途

一套完全自动化的新闻文章生成管道，通过定义的自治阶梯，从阴影模式开始，最终在满足严格质量指标后实现完全自治发布。

问题

How to safely let an LLM publish content to production without human oversight while ensuring quality and trust.

结果

The pipeline achieved $2.26 cost per article, 0 banned phrases, 1.0 structure score, 0.98 groundedness, and maintained a 4.0 mean persona adherence. It moved from shadow to autonomous in eight weeks, publishing 69 articles weekly without human intervention.

工作流

• 1. Pull weekly source articles from feeds
• 2. Cluster similar stories and deduplicate
• 3. Run novelty check to filter duplicates
• 4. Cap topics to a limited set
• 5. Rewrite 69 articles using four editorial voices
• 6. Generate images for each article
• 7. Write rewritten content and images to production database
• 8. Trigger site rebuild and search index refresh
• 9. Log all outputs
• 10. Run mechanical quality checks (structure, groundedness, banned phrases)
• 11. Sample articles for claim verification by a Haiku judge
• 12. Score articles with a multi‑judge persona panel
• 13. Store scorecard and telemetry in blob storage
• 14. If all metrics meet thresholds, promote to next rung; otherwise demote

证据

If you'd asked me eighteen months ago whether I would let a language model publish to production with no human in the loop, I would have said no

来源

The summary landed in blob storage next to the telemetry: banned phrases 0, structure 1.0, groundedness 0.98, and cost $2.26.

来源

The rewrite stage of my pipeline is on rung four.

来源