When AI Agents Meet Vision Therapy: A Builder's Personal Take on Amblyotube

Seeing Momentic hit the top of Hacker News this week — AI agents that catch bugs by actually using your app like a human — felt like a flash of recognition. 143 points, 95 comments. Developers instantly grasped the insight: the best automation doesn't replace human judgment, it extends it.

I've been living that principle for the past year, but in a completely different domain. At Seven Sports, we asked: what if an intelligent layer that works with human perception could help people with amblyopia retrain their vision?

The result is Amblyotube on Meta Quest. You put on the headset, pull up any YouTube video, and our dual-eye shader pipeline presents a slightly different image to each eye: adjustable blur on the dominant eye (digital occlusion), AI-powered figure highlighting via MediaPipe on the weaker eye (MFBF mode), optional wedge occlusion to reduce binocular rivalry, and a magenta fusion cue that teaches the brain to merge both views into true 3D depth.

The engineering challenge was making this feel invisible. 30–40 minutes a day, content you already love, no patches, no red-blue glasses. The LogMAR vision check built into the app uses a clinical staircase algorithm with Landolt C optotypes at virtual 3m — it auto-calibrates your therapy profile each session, adjusting blur intensity and highlight strength based on your measured acuity.

If you're building at the intersection of perception and software, this is worth studying.

👉 https://www.meta.com/en-gb/experiences/amblyotube/25906906972338493/