Nate B Jones discusses the different assistants he uses and argues that a more capable model cannot necessarily overcome a website that has chosen to restrict agents. He describes the difficulty of distinguishing authorised customer automation from malicious traffic.
Nate B Jones examines examples of signed agent requestsAn AI agent identity is a persistent, verifiable identifier used to attribute messages, actions, permissions, and history to an agent., payment controls and merchant information. He presents these as developing approaches, with businesses still deciding which agents and actions they will accept.
The episode proposes three broad access situations: formal partnerships, agents navigating ordinary websites, and sites refusing agent access. Nate B Jones also discusses prompt injectionPrompt injection occurs when untrusted content attempts to redirect an AI system's behaviour beyond the authority of that content. and encourages matching delegated tasksAI agent delegation assigns a bounded task, context, authority, and expected result from one participant to an agent or another agent. to an assistant’s actual access and reliability.
Watch on YouTube
