NLNathan Labenz@labenz8:58 AM · Aug 20, 2026

1496AI

☀️ AI:AM is live. Today: Matt Harpe + @mitch_troy of @trybasis on running long-horizon agents on real money, 9:30a PT. Then Jay Dawani, co-founder & CEO of Lemurian Labs, on whether CUDA's moat is really a software problem, 10:15a PT.… Show more

NLNathan Labenz@labenz9:05 AM · Aug 20, 2026

1496AI

☀️ AI:AM is live. Today: @mitch_troy, co-founder of @trybasis, on running long-horizon agents on real money, 9:30a PT. Then Jay Dawani, co-founder & CEO of Lemurian Labs, on whether CUDA's moat is really a software problem, 10:15a PT.… Show more

NLNathan Labenz@labenz9:29 AM · Aug 20, 2026

1496AI

Reacting to a robot demo on AI:AM — Prakash: "I was hoping robotics would take a little bit longer." China is way ahead on the physical side; the brains are coming along. Nathan: roughly a GPT-3 moment — vs. the consensus, he'd probably take the under on timelines.

NLNathan Labenz@labenz9:35 AM · Aug 20, 2026

1496AI

The old robot benchmark: can it come into your home and make you a cup of coffee? Nathan's revision, floated on AI:AM: "can it do it if you show it how to do it a couple times?" His caveat: 99% reliability might cut it for code — maybe not in your kitchen.

NLNathan Labenz@labenz10:03 AM · Aug 20, 2026

1496AI

How Basis supervises runs that last eight hours, sometimes longer (@mitch_troy on AI:AM): not token-level process rewards — closer to supervising a human. Did you do the step? His example: if the agent made a PowerPoint, did it visually render the images to check?