What it is.
Six workflows, each with its output shown: a listicle builder, a long-form drafter, content refresh, citation gap outreach, an SEO brief generator, and an image and screenshot pipeline.
Brief approval runs 2 to 3 minutes each and final review 8-12 minutes per article, against about 3 hours per article before the system. Three caveats open the piece, including that none of it is for scaling low-quality content.
Slate's own account of building on Slate, so running it as described needs their platform. The six workflow shapes and the review accounting transfer anywhere.
What you get.
- Six named workflows with the output each produces.
- Human time accounted for per stage, at 8-12 minutes of final review per article.
- The before figure, at about 3 hours per article with earlier tooling.
- A distinction between workflows and agents, where an agent is a workflow with memory.
- Three stated caveats, including that topic selection still matters more than the system.
How to set it up.
Pick the one workflow with the most repetition in your own process and build only that.
Count your current human time per article before you automate, so you have a baseline.
Build the brief generator before the drafter, since a bad brief makes a bad draft faster.
Keep a review gate with a stated time budget, at roughly 8-12 minutes per article here.
Add memory to a workflow only once it runs reliably without it.
Treat topic selection as a human decision that sits outside the system.
Use cases
See the whole pipeline
Read the six workflows and what each one produces before building any.
Budget the human time
Use the per-stage review minutes when planning capacity.
Pick the first workflow to build
Start with the stage that costs your team the most time today.
Best for
Deciding what to automate first
Six workflows with outputs shown is more useful than a list of possibilities.
Costing a content system
Per-stage human minutes is the number a business case needs.
Arguing against volume automation
The caveats section says outright that this is not for scaling low-quality output.