Notes
Short notes on AI-native product design.
Custom Eval Infrastructure at With Him
How With Him uses a first-party policy engine, shadow traffic, and launch gates to keep assistant behavior tied to product policy.
Building Trustworthy AI Memory at With Him
Summarized continuity, user control, and bounded personalization, how With Him implements memory as a policy-gated pipeline instead of raw retrieval.
Most AI Products Optimize for Intelligence Instead of Trust
The gap between technically impressive AI and AI that users actually return to often comes down to one thing: trust architecture. Teams build for capability when they should build for reliability.
The First 60 Seconds of an AI Product Matter More Than Teams Think
First impressions in AI products set the frame for everything that follows. Most teams underestimate this window and lose users before the product can demonstrate its value.
Memory in AI Products Is a Trust Contract
When an AI remembers something about a user, it makes an implicit promise. Breaking that promise, forgetting inconsistently, or remembering inappropriately, damages trust more than never remembering at all.
AI Agents Are Most Valuable When They Reduce Operational Drag
The best AI agents don't replace human capability, they handle the execution that humans deprioritize. Find work that users avoid, not work they can't do.
Great AI UX Helps Users Form Better Prompts
Prompt engineering shouldn't be the user's job. Great AI products guide users toward better inputs through design, not documentation.