Blog
Notes on AI-managed ad spend
How the guardrails, the attribution window and the learning loop work in practice for apps, websites and businesses — written from the code, not from a slide deck.
- Attribution · 6 min read
Why one change per campaign every 72 hours
Attribution is late, and every network is late differently. That is why the agent is not allowed to react twice.
Read article - Safety · 7 min read
Guardrails are code, not a polite request in a prompt
A model that is allowed to move money needs limits it cannot argue with. Here is the list we enforce before any write.
Read article - Data · 5 min read
Four networks, four definitions of “an install”
The same campaign can look profitable in one console and unprofitable in another. The difference is usually semantics, not performance.
Read article