The part of my AI content engine that surprises people most isn't the writing. It's the self-learning loop. Every Saturday I download the week's analytics from Typefully, upload the CSV to Claude, and ask it to find patterns. What drove engagement? What fell flat? Why? Claude analyses the data, extracts specific lessons, and saves them into Notion. The next week, when it generates content, it references those lessons automatically. Week 1: content is good. Week 2: content is better because it learned from Week 1. Week 4: the engine is producing insights I wouldn't have spotted manually. The Japanese call this Kaizen. Continuous improvement. 1% better every week. Last week's Kaizen lesson: AI crossover posts get 10-50x the reach of crypto-only posts. So the engine started weighting them higher. That's not me making a decision. That's the system optimising itself. $30/month for Claude. $20/month for Notion. That's the whole stack.