SpaceX Engineer: Always-On AI Agents Hit a 200K Token Wall

A SpaceX engineer's Stanford lecture shows why AI agents stall at a 200,000-token memory wall, and why the model isn't the bottleneck. The harness around it now decides whether agents can work unattended for weeks.

Chapters

(0:00) AN ALWAYS-ON

(2:18) The architectural shift toward ambient

(4:44) The isolated cloud computer

(5:25) System Blueprint Step 3: Surviving the

(8:09) Structural

(8:48) Designing the cognitive memory mode1

(10:05) 10to 60 Minutes

(10:44) Designing the cognitive memory mode1

(12:02) Structural

(13:21) The Escalation Ladder

(15:06) System Diagnostics

(16:17) Assume the models today are

(17:54) System Diagnostics

(21:18) Apps for Intelligent

(21:52) Get the free apps

DeepValues toolkit

• Free Margin of Safety Weekly, every Saturday (guru moves, market temperature, the week's research; no account needed)

• Open-source research

• DeepValues for iOS, intrinsic value and margin of safety for any stock

• Is the market overvalued? Track it with Buffett Indicator (iOS)

• The same market gauge on Android, Buffett Indicator & Shiller PE

• Deep Value Intelligence, the multi-agent AI analyst

• Developer API for your AI agents

Deep Values explains AI from first principles: the research papers, models, chips and companies shaping the AI era, and what they mean for builders and investors. Understand AI from first principles.

• This episode on YouTube

Subscribe

Podcast (Deep Values Frontier): Spotify · Apple Podcasts

Investing channel