An AI Operating System
summary
Claim: treat AI as an operating system by centralizing usage and measuring real impact. Built Aperture to proxy models, enforce identity-based access, log calls and show hundreds of millions of cached tokens for a single engineer. Shows Claude Opus 4.6 as the heavy spender, reroutes repeatable work to cheaper models, turns raw token bills into operational metrics, and uses usage data to justify more headcount while cutting model costs.
AI-written summary · about
chapters (12)
- 0:00AI operating systems intro
- 1:05Automation to deep reasoning
- 3:34AI arrival and challenges
- 6:01Measuring AI systems
- 8:08Centralize workflows with Aperture
- 10:57AI usage metrics and insights
- 13:29High performers and token efficiency
- 16:01Identity-based AI access
- 17:45Route to cost-effective models
- 18:49Token spend insights
- 20:12AI changing leadership
- 21:16Future of AI productivity
related talks
32mBuilding an AI Operating System
38mAI Engineering Stack: Assisted to Delegated
24mPlatform Engineering and AI - Two Buzzwords Finally Meet!
33mPowering AI Capabilities with API Management
35mCanal Mania and the Philosopher's Stone
33mSolving business challenges in Gen AI with Anthropic
22mWhen AI Agents Start Acting, APIs Become Execution Contracts
31mAI APIs: Taming the Wild West with AI Gateways