It isn't a quality problem—it's a translation problem: qualitative insights don't become backlog items because there's no formal process to bridge the formats. The fix is to embed decision-makers in synthesis, translate findings into prioritization-ready format (impact + effort), and track their status like bugs—otherwise, research remains an expensive activity with no measurable value

AI can help organize existing research into empathy maps, but it cannot replace real user evidence—because plausible, generic quotes (e.g., "I wish the app would ask before swapping") are not the same as specific, messy details from real sessions (e.g., "It swapped my oat milk for whole milk and charged me before I saw the notification"). Before reaching for AI, ask: am I organizing real data or generating data I don't have? If the answer is the latter, the map needs more research, not more AI

The UK emergency alert about wildfires was well-intentioned but failed its users: dense text, redundant language, weak visual hierarchy, and unhelpful signposting made it harder to scan in a moment of stress. A content redesign—shorter sentences, direct "do not" commands, clear paragraphs, and a specific URL—cut reading time from 19 to 14 seconds and dropped the reading age from 17.6 to 14.6, proving that in an emergency, every word and second counts

When a stakeholder says "we need a survey by Tuesday," the real skill isn't asking for more time—it's starting with the question, not the method, and doing credible research within imperfect conditions. Fast research means reducing scope, focusing on the most critical questions, and making deliberate trade-offs (e.g., 5 relevant users instead of 20 later)—and sometimes the method itself evolves as findings emerge, because research is about following the evidence, not sticking to a plan

Craft isn't about pixels anymore—when AI makes production cheap, real craft moves to choosing the right problem and curating AI output with taste. To keep it, write your standards down (so the machine applies your rules), keep research a weekly habit, and always put a human review at the end—because AI fails quietly, and fluent confidence without verification is just fiction

People ignore perfectly designed buttons not because of color, size, or placement, but because of the trust problem (uncertainty about what happens next), decision fatigue (too many choices already made), and identity questions (not ready to be the person who clicks it). The fix isn't polishing the button—it's building the right emotional state, trust, and narrative momentum before it appears, because a button is just the answer to a question the experience needs to ask first