Build a newsletter-to-social content agent (it failed)
The author built a content agent to research, write and schedule social posts, but it invented statistics, so they rebuilt it as a structured workflow.
What they did
The author spent three weeks on one agent that had a content library, brand voice guidelines, engagement metrics and trend-research tools. It was meant to turn newsletters into social posts. After it failed, they rebuilt it as a fixed eight-step workflow. The steps were insight extraction, research, synthesis, hook, structure, conclusion and voice check. Each step had guardrails, and a human review gate came at the end.
How it went
The agent invented a "73% of knowledge workers" statistic and blended two newsletter topics into one post. Cleanup took about three hours for each hour it saved. The workflow gave consistent output, and the author could fix one step at a time. The post gives no metrics for the new version.
Worth knowing
The author refined the workflow eight times, so expect iteration. Reviewing research quality before the hook step catches the most damage.
Labels
It was meant to turn newsletters into social posts.
The agent invented a "73% of knowledge workers" statistic and blended two newsletter topics into one post.
After it failed, they rebuilt it as a fixed eight-step workflow.
The author built a content agent to research, write and schedule social posts, but it invented statistics, so they rebuilt it as a structured workflow.
The “Material” label failed its quality check and is not used for selection.
Details
- Agents
- Zapier Agents
- Source
- blog
- Source date
- undated
- Accounts
- 1
- In registry since
- Oct 12, 2026