Social Media Autopilot
Five agents that write, critique and publish platform-specific posts. Then I measured whether they beat one good prompt.
The problem
Everyone claims multi-agent pipelines write better content than a single good prompt. Almost nobody checks.
What I built
You give it a brief. Five agents write and critique posts for LinkedIn, X, Instagram and Reddit. You approve, it schedules and publishes.
Then I scored it properly: myself plus five reviewers rated both approaches across five dimensions. I also benchmarked three local models and threw one out, because GPT-OSS-20B ran fine technically but shipped inconsistent output.
How it went
The pipeline scored 4.1 out of 5. A single prompt scored 3.1. Publishing was tested end to end on my own accounts, not simulated.
Numbers
- 4.1 / 5
- 5