Frameworks, core principles and top case studies for SaaS pricing, learnt and refined over 28+ years of SaaS-monetization experience.
Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.
.png)
The Claude Code leak produced a fascinating natural experiment in defensibility. Within hours of the source code being archived, developer Sigrid Jin had created a clean-room rewrite of the core architecture in Python, using a tool called oh-my-codex. The repository, called claw-code, hit 50,000 GitHub stars in two hours, one of the fastest-growing repositories in the platform’s history.
This would seem to undermine the argument that the harness is defensible. If the architecture can be reverse-engineered and reimplemented overnight, where is the moat?
The answer lies in the distinction between architecture and tuning. Knowing the architecture of Claude Code tells you the shape of the system. Its components, their relationships, the general approach to orchestration and memory and coordination. It is like having the blueprints for a Formula 1 car. The blueprints are valuable, but the car that wins races is the one whose every component has been tuned, tested, and re-tuned over thousands of laps on specific tracks with specific drivers under specific conditions. The tolerances, the edge cases, and the institutional knowledge about which configurations work and which ones cause subtle failures under load. All of it matters. A clean-room rewrite of Claude Code’s architecture gives you a starting point. It does not give you the thousands of hours of tuning that make the orchestration work well with Opus 4.5’s specific behavioral patterns.
More importantly, it does not give you the feedback loop. Anthropic can use data from Claude Code’s usage to improve both the harness and the model simultaneously. The harness team discovers that the model struggles with a certain class of orchestration instructions. The model team adjusts post-training to handle those instructions better. The model team releases a new capability, and the harness team immediately builds orchestration patterns that exploit it. This co-evolutionary dynamic is the true source of durable advantage, and it is not something that can be replicated by reading leaked source code.
For anyone pricing an agentic AI product, the lesson is this: your defensibility is not in your architecture (which can be reverse-engineered), nor in your model (which will be matched by competitors on benchmarks). It is in the depth and velocity of your integration between the two. Price accordingly.
Join companies like Zoom, DocuSign, and Twilio using our systematic pricing approach to increase revenue by 12-40% year-over-year.