AI governance
Four eyes on every post: a shared governance foundation
Starting questionHow do we make sure AI-assisted social media posts never go live without human approval, across several channels?
Result
Works
A shared approval library serves several channels. Each channel still enforces four-eyes approval itself.
Through one shared, versioned approval library instead of copied logic per channel. A central controller distributes the posts, yet every channel enforces four-eyes approval itself. If the content changes, a hash invalidates the approval.
What we built
We started with a server for one channel, with four-eyes approval, scheduling, and a log. For the second channel we moved the approval logic into a shared, versioned library instead of copying it.
A central controller distributes posts across channels. Each channel still enforces approval on its own.
What it means for business architecture
Approval rules are a shared capability of the company, not a feature of a single channel. Copy them several times and you end up with diverging rules.
Lessons learned
- Copied approval code inevitably drifts apart over time.
- An approval becomes invalid as soon as the content changes. The only reliable way is a hash of the content.
- A shared foundation needs its own release process. That takes effort, and it pays off.
- four-eyes principle
- approval
- reuse
- social media