42:I[185,["/_next/static/chunks/3xnt26d00_zk-.js","/_next/static/chunks/1s-yni1uf2lwd.js","/_next/static/chunks/42spz6paskvb6.js","/_next/static/chunks/06x30xgfe6iur.js","/_next/static/chunks/3b4qrzgkt_l12.js","/_next/static/chunks/15lgs8s22dk5l.js"],"SendEmailButton"] 44:I[63625,["/_next/static/chunks/3xnt26d00_zk-.js","/_next/static/chunks/1s-yni1uf2lwd.js","/_next/static/chunks/42spz6paskvb6.js","/_next/static/chunks/06x30xgfe6iur.js","/_next/static/chunks/3b4qrzgkt_l12.js","/_next/static/chunks/15lgs8s22dk5l.js"],"AddToLibraryButton"] 45:I[29721,["/_next/static/chunks/3xnt26d00_zk-.js","/_next/static/chunks/1s-yni1uf2lwd.js","/_next/static/chunks/42spz6paskvb6.js","/_next/static/chunks/06x30xgfe6iur.js","/_next/static/chunks/3b4qrzgkt_l12.js","/_next/static/chunks/15lgs8s22dk5l.js"],"ShareButton"] 46:I[58016,["/_next/static/chunks/3xnt26d00_zk-.js","/_next/static/chunks/1s-yni1uf2lwd.js","/_next/static/chunks/42spz6paskvb6.js","/_next/static/chunks/06x30xgfe6iur.js","/_next/static/chunks/3b4qrzgkt_l12.js","/_next/static/chunks/15lgs8s22dk5l.js"],"ProcessEpisodeButton"] 47:I[45269,["/_next/static/chunks/3xnt26d00_zk-.js","/_next/static/chunks/1s-yni1uf2lwd.js","/_next/static/chunks/42spz6paskvb6.js","/_next/static/chunks/06x30xgfe6iur.js","/_next/static/chunks/3b4qrzgkt_l12.js","/_next/static/chunks/15lgs8s22dk5l.js"],"SignupCTA"] 3a:["$","time",null,{"dateTime":"2026-05-25T05:20:38.000Z","children":"May 25, 2026"}] 3b:["$","span",null,{"aria-hidden":true,"children":"·"}] 3c:["$","span",null,{"children":"00:15:31"}] 43:T50a,

Every team that won the Gemini 3 Hackathon built a workaround for something AI can't do. That's not a coincidence — it's the pitch, read backward.

I looked at all three winning products: a supply chain crisis tool, a disaster triage system, and an assistive navigation app. Each one won by designing around an AI failure — and each workaround comes with a cost the sales deck doesn't mention.

More of this in the newsletter → [2nd Order Thinkers URL]

What you'll leave with: A framework for reading any AI pitch in reverse. Stop at the human approval gate, the explainability layer, the memory patch — and ask what it's compensating for. That question is worth more than any demo.

Mentioned in this episode:

Globot (Gemini 3 Hackathon grand prize)

Aegis (2nd prize)

Netra / Memory Palace (3rd prize)

Deloitte / UK Ministry of Defence automation bias report

Carnegie Mellon + Stanford research on human-agent supervision costs

This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit www.2ndorderthinkers.com/subscribe