Suganthan watched Claude pay a penny to read his website

Suganthan built a demo where his personal website charges AI agents one cent per page view. Claude paid the fee and retrieved the content on its own, with no human stepping in to approve the transaction. The demo is small, but the behavior is new: an agent spending real money to complete a task. This opens a category of spending that current agent tools have not addressed. Most agent products let you set a budget for compute or software service calls. Paying third-party vendors mid-task, in small per-action amounts, is a different thing. The agent held enough authority to spend money on the human's behalf. The human set no ceiling, saw no confirmation request, and had no record of the spend until after it happened. For product teams, the design gap is concrete. A human needs to know, before handing a task over, what the agent is allowed to spend and with whom. After the task, the human needs a readable record of what was paid and why. Neither of those interactions exists in mainstream agent products today. The demo is a small proof of concept, but it shows a class of permission that will need its own design answer.