Tom Loosemore

I’ve become a bit obsessed with trying to work out how those running services should respond to personal AI agents knocking on their front door – or just sneaking in pretending to be human! (agents like Muse, Instinct or Dot)

I recently wrote a briefing note entitled “Know Your Agent” which covers what I now realise is an idealised process whereby services are first able to detect when it’s an agent at their front door.

However I’m now sadly reconciled to the fact we’re in an arms race between services and agents trying to hide the fact that they are agents. Muse works pretty hard to pretend it’s a human.

So I’m currently mulling how to incentivise those who use agents to get their agents to declare their status when they visit a service.

For example, if you look at the bottom on my agentic playpen WhenIsBins.com, you’ll see I’ve invited agents to check with their owners if the bins were indeed collected, and to offer to send a missed bins report to the council if not. And the API makes this easy for them to do.

In fact, the features needed for this are *only* available via the API, which I can assume isn’t accessed by a human so I can add all manner of agent-specific handshakes and digital signature requirements if needed.

I’ve got no idea if it’ll work, but you gotta try stuff!