Daily Dilemma is an AI-first product experiment. Every part of the software delivery lifecycle, from content to analysis to optimization, has an AI layer to empower decision-making. The final decision always rests with a human, but AI does most of the heavy lifting.
Every week, Claude generates 7 new dilemma questions — fun, polarizing, non-controversial. They're scheduled one per day and stored in a database queue.
Anonymous analytics track how people interact: pageviews, votes, time to vote, shares, and return visits. No personal data is collected — just anonymous session IDs.
A weekly scan feeds the analytics into Claude, which identifies UX issues and opportunities. It outputs ranked hypotheses formatted as “We believe [change] will [outcome] measurable by [metric].” A daily health check tracks metrics against baselines and flags anomalies, keeping the feedback loop tight without waiting a full week to react.
Numbers only tell part of the story. Real user feedback submitted through the app is reviewed alongside the data so human voices shape what gets prioritized, not just metrics.
Hypotheses get turned into A/B tests via feature flags. A daily pipeline coordinates the experiment lifecycle: it checks dependencies, recommends the next hypothesis to test, and an experiment monitor evaluates statistical significance to decide whether to ship, roll back, or keep running.
A watchdog runs every hour. If error rates spike, it automatically disables all feature flags and sends an alert. Experiments that cause problems get killed before they cause real damage.
We believe adding a like button to dilemmas will increase engagement, measurable by likes per vote and return rate
running · 2026-02-17 · via manual
We believe adding a skip button to dilemmas will increase vote completion rate, measurable by vote_cast / pageview ratio
proposed · 2026-02-17 · via manual
We believe a Top 10 most liked dilemmas page will increase return rate, measurable by sessions with >1 pageview
proposed · 2026-02-17 · via manual
We believe showing a teaser of the next dilemma on the results screen will increase the rate of users answering a second dilemma, measurable by keep_going_clicked / results_viewed
running · 2026-02-25 · via manual
We believe feeding like/skip signals into dilemma generation will improve content quality over time, measurable by like ratio and vote completion rate trending upward
proposed · 2026-02-17 · via manual
Started 2026-02-25
Started 2026-02-17