Episode Summary:
Brandt and Will trade war stories on the gap between AI hype and AI reality: Brandt's saga of nursing a decade-old, server-killed Oculus Go back to life with Claude Code, the "caveman concept" for cutting token waste, and why Claude keeps giving up and trying to open Chrome. Will breaks down Anthropic's new (and deliberately withheld) Mythos model, plus his own iMessage-agent setup via BlueBubbles. They wrap with confidently-wrong AI moments and this week's Things of the Episode.
Discussions Include:
• Claude Cowork vs. Claude Code vs. Claude chat, and when to reach for each
• The "caveman concept": talking to your AI simply to burn fewer tokens
• Resurrecting a 10-year-old, server-bricked Oculus Go using Claude Code, ADB, and a custom APK
• Anthropic's Mythos model, its bug-hunting/exploit-chaining chops, and Project Glasswing
• When AI is confidently wrong (and how to stop it defaulting to "just open Chrome")
Quotable Quotes (Should you choose to share):
"It is so stupid sometimes... but at the same time, it is also still really, REALLY stupid sometimes." - Brandt Krueger
"There's two types of people in this world: people who use AI to learn everything, and people who use AI to learn nothing." - Will Curran
"It flippin' writes its own APK... " - Brandt Krueger
"Definitely the power is on the code side of things. It's much smarter." - Will Curran
Thing of the Episode (TOTE):
Brandt: Oculus Go hacking project - https://www.ebay.com/b/Oculus-Go-VR-Headsets/183068/bn_7115189259
Will: BlueBubbles - https://bluebubbles.app/