When I left my job at WIRED, one of the first things I did was grant access to my email and calendar to an AI agent called Cowork. My expectations were high, but reality was more modest. Cowork turned out to be just a fancy chatbot, albeit with some handy tools.

Then Instinct, an exclusive AI agent, caught my attention. It offers a more seamless experience, connecting with my email, calendar, and messaging apps. While Meta’s Muse became the most downloaded free app, its launch was marred by security vulnerabilities. I remain cautious, but if Instinct could handle tasks like unsubscribing from a persistent email list, I might reconsider.

The divide between those who embrace AI agents and those who are wary is growing. For tech CEOs, the benefits of automating administrative tasks seem worth the data centre costs. For others, the deal might not be as attractive. The challenge, as technology journalist Jasmine Sun notes, is that most people’s problems are not software-shaped.

I decided to give Instinct a try. With the right form factor, it excelled at booking reservations and even handled a ticket cancellation, much to my surprise. However, the risks are significant. Security breaches, data misuse, and even minor errors can lead to frustration. As I navigate these challenges, I can’t help but feel the AGI is closing in.