The desire to push back into MCP highlights one of the primary tensions with AI agents in my mind. For an AI agent to be useful it needs to have broad access. For an AI agent to be safe it either needs limited access or perfect alignment. And perfect alignment doesn't currently exist.
Comments
The desire to push back into MCP highlights one of the primary tensions with AI agents in my mind. For an AI agent to be useful it needs to have broad access. For an AI agent to be safe it either needs limited access or perfect alignment. And perfect alignment doesn't currently exist.