How Anthropic Safely Ships the World’s Best LLMs: Decoupling the “Brain” from the “Hands”
Katelyn Lesse and Angela Jiang join Latent Space to unpack the engineering behind the Claude platform: shipping Opus 5.5, decoupling the model’s “brain” from the agent’s “hands”, scaling the API, and building reliable Managed Agents.
A newly recorded intro covers Anthropic’s S-1 filing, improvement and investment flywheels, and why the platform matters.
Watch the full episode on YouTube
Show Notes
- Claude Opus 5.5 — Anthropic’s release notes on model behavior, evaluations, and pricing.
- Scaling Managed Agents — Decoupling the brain, harness, and sandbox.
- How Anthropic contains Claude — Isolation, credentials, and network controls.
- Effective harnesses for long-running agents — Maintaining task momentum across context windows.
- Anthropic S-1 Breakdown — Context for the newly recorded intro.
Timestamps
0:00 Anthropic’s S-1 & This Interview
1:06 Revenue per Employee & Efficiency
3:13 Improvement Flywheels & Investment Flywheels
4:04 Why the Claude Platform Matters
6:25 Training Incidents & Production Guardrails
7:30 Opus 5.5 Launch & Pricing
9:25 Distillation & Preserved Thinking
14:35 Product & Engineering Organization at Anthropic
18:05 Scaling the Claude Platform
19:25 Model Access & Safety Decisions
26:25 Trust Grants & Product Roadmap
30:25 How Anthropic Ships Models
35:55 Benchmarking & Model Personality
42:10 Claude Managed Agents
44:25 Memory, Dreaming, & Agent Harnesses
48:25 Evaluating Agents
53:25 CPU Bottlenecks & Scaling Out
56:05 Rewriting the Messages API in Rust
1:03:25 The Brain vs. The Hands
1:05:25 Securing Credentials & Sandboxes
1:08:25 Training Incidents vs. Production Incidents
1:10:55 Frontier Model Cadence & Releases
1:12:25 Safety Classifiers & Model Routing
1:19:25 Emerging Agent Workloads
1:20:55 Combating AI Slop
1:27:25 Management, Writing, & Work Styles
1:33:25 Building Anthropic’s DevRel Team
1:40:25 DevRel & Having a Product Voice
1:43:25 What to Build with Opus 5.5