Platforms are fencing in AI agents

A post on Apple and a hacker's future argues Apple is unhappy about agents, after making Full Disk Access more explicit. Commenters split. One said it's just a security prompt. Another said it is about platform control: Apple limits background autonomy while first-party frameworks keep ambient access. Separately, Anthropic told Claude Code subscribers they can no longer use OpenClaw, citing 'outsized strain', with a one-time credit as a sweetener. Users answered that using the capacity you paid for is not abuse, and asked for capped API keys instead.

The pattern: whoever owns the OS or the model subscription treats third-party agent harnesses as a threat to margins and control. Even the runtime layer is part of it. Commenters on a Node vs Deno post mentioned Bun's Anthropic acquisition and Rust rewrite drama as a reason it's no longer a fan favorite.

The counterpoint is that these are real security and capacity issues, and the rules were stated in advance. But the direction is consistent: agents are welcome when they run inside the owner's product.