Thinking
Field notes from the work — what generalizes, what breaks, and what holds up.
What it takes to make an AI reporting pipeline inspectable end to end — collection, clustering, drafts, publishing, and traces.
Read ↗
AI agents lose context and repeat mistakes. memory-os treats memory as a governed artifact instead of a scrap pile.
Delivery is the unglamorous middle between a proposal and a live site. Two projects — a furniture brand and an institutional publisher — show what holds up.
Speed is the easy promise of automation. Reviewability is what makes it safe to run. Two messaging and content automations show the pattern.