diff --git a/ai_preface.asciidoc b/ai_preface.asciidoc
index 18ba3839..76b114bb 100644
--- a/ai_preface.asciidoc
+++ b/ai_preface.asciidoc
@@ -97,6 +97,10 @@ Its ability to write that _first_ test,
 the one where, as we'll see, a lot of the design (and thinking) happens in TDD,
 was much more mixed.
 
+// SEBASTIAN: Idea for a frame with actionable advice: "Make sure there are at least few examples of tests before you'll make AI assistant write more. Show it what 'good' looks like."
+
+// SEBASTIAN: Another idea for a frame with actionable advice: "Simple things like writing a meaningful test name or describing in comment what you want to test can help immensely AI working in autocomplete mode."
+
 Similarly when working in a less "autocomplete" and more "agentic" mode,
 I saw AI tools do very well on simple problems with clear instructions,
 but when trying to deal with more complex logic and requirements with ambiguity,