TL;DR.
The Harness Thesis: How Anthropic Proved the Model Doesn't Matter, Then Lost Control of What Does
The abstraction ladder
AI tooling evolved from 6-layer SDK wrappers to zero-code configuration in 18 months. Each rung assumes greater model competence and reduces developer scaffolding.
Ten features in ten weeks
Anthropic shipped Opus 4.6, Agent Teams, Auto-Memory, Code Review, App Store, Academy, and Xcode MCP — proving that what wraps the model is where value accrues.
The C&D that changed everything
Anthropic's cease-and-desist forced a rename and pushed OpenClaw from Claude-native to model-agnostic. OpenAI hired the creator two weeks later. Jensen Huang name-dropped the project at GTC.
The domain gradient
Coding succeeded first because verification is automated. Semi-formal and situated domains — where 89% of enterprise knowledge lives — need configurable platforms accessible to non-developers.
The commercialization gap
68% of enterprises are stuck in pilot phase. Only 6% of CIOs consider their orgs AI-ready. The bottleneck isn't model quality — it's the gap between model capability and domain expert accessibility.
Quick Synthesis
“Anthropic built the intellectual framework for understanding that value migrates from model to harness. Then, when a third-party project appeared that would extend their harness thesis to the 89% of enterprise knowledge that's currently uncaptured, they sent lawyers instead of engineers.”
Generated by Intelligence Unit
AI-synthesized from article content