Can AI actually beat prediction markets — or does the house always win?

In Episode 26 of Built This Week, Sam Nadler and Jordan Metzner kick off 2026 by breaking down a real AI trading bot Jordan built for prediction markets like Kalshi, using live market data, whale detection, coordination signals, and confidence scoring.

Jordan walks through the full system — backend, frontend, live alerts, and execution logic — and shares the honest results: a 66 percent win rate that still lost money once fees and market dynamics were factored in.

The takeaway is not hype. It is reality.

The episode also dives into:

• Why prediction markets feel like gambling but are regulated differently
• How insider-like signals emerge from coordination and volume behavior
• Why bots end up trading against bots
• Where real alpha might exist (and where it does not)

We also cover:

• Google NotebookLM as a serious education and onboarding tool
• Turning documents into infographics, slide decks, and audio learning
• Nvidia entering autonomous driving and competing with Tesla
• Nvidia’s new Rubin architecture and why it matters
• Tesla vs Waymo economics and the future of Full Self Driving
• Why Anthropic and Claude Code are becoming developer defaults

This is not theory.
This is what happens when you actually deploy AI systems into real markets.

Timestamps

(0:00) Why prediction markets are exploding
(1:07) Episode 26 kickoff
(2:00) Why build a trading bot at all
(4:30) Kalshi vs Polymarket APIs
(6:00) Live market signals and whale detection
(9:30) Win rate vs profitability
(12:00) Why fees destroy returns
(14:30) Bots trading against bots
(17:00) Where real alpha might exist
(18:00) NotebookLM for learning and onboarding
(21:00) Nvidia enters autonomous driving
(24:00) Tesla vs Waymo economics
(27:00) Nvidia Rubin chips explained
(28:30) Anthropic and Claude Code momentum
(29:30) Final thoughts

Links

Built This Week
New episodes every Friday

Jordan Metzner
https://x.com/mrjmetz

Sam Nadler
https://x.com/Gravino05