Claude, OpenCode, and the Death of “Bring Your Own Harness”

There is a lot of noise about Anthropic censoring the word “OpenCode”. I saw a tweet from Ryan Vogel showing that they might be blocking “OpenCode” specifically in system prompts. I cannot verify if this is universally true, but if they are flagging the keyword to terminate API connections, it is a technical block to prevent unauthorized client access, not necessarily a ban on discussing the topic.

Anthropic is closing an economic loophole. Users were using the Claude Code Max subscription to run expensive autonomous tasks that would cost significantly more through the standard API. From a business perspective, subsidizing competitors’ tools at a loss is unsustainable.

However, this move signals a larger trend: Anthropic is becoming the Apple of AI. They are building a walled garden where everything happens behind closed doors. They don’t just want to sell you the model; they want to control the entire experience. If you don’t use their specific interface, you don’t get to play. This isolationism is the opposite of what drove the initial explosion of AI utility.

The market responded immediately. OpenCode didn’t just accept the block; they pivoted. They announced partnerships with other providers who are willing to support their ecosystem and launched OpenCode Black, their own subscription service. The tool is evolving to bypass reliance on a single, hostile provider.

This shift confirms why I moved away from Claude after the Sonnet 3.5 and 3.7 releases. I now rely on Gemini, Codex, and open models like Qwen Code and GLM. These alternatives integrate with the tools I actually use. Anthropic may protect its margins by blocking third-party harnesses, but the rest of the ecosystem is moving on without them.