Testing an AI Agent That Mostly Delivers Despite Security Flaws
The author tested Instinct, an invite-only AI agent that connects to email, calendar, and messaging apps. It saved money, booked reservations, and flagged a phishing attempt, but also wasted some money and raised security concerns. The piece discusses the growing divide between people who use AI agents and those who don't, and the backlash against data centers.
Instinct is an invitation-only agent that works through iMessage and WhatsApp and links to a user's email, calendar, and messaging services. It entered private beta in February, and reports say it is negotiating a $1 billion round after already raising $350 million, which would value it at $10 billion.
In testing, it saved $550, booked restaurant reservations, and identified a phishing attempt, but also wasted $64 and raised security worries. Its design avoids an open chat box, offering prompts instead; unlike Lindy, which texted daily and joined Zoom calls uninvited, it reportedly does not over-message.
As AI agents like Instinct become more capable, they may deepen a split between people who delegate email, scheduling, and errands to software and those who do not. Early adopters could gain time and savings, while others may see fewer benefits and more disruption from the infrastructure powering these tools. Security incidents and wasted spending may also shape trust, influencing whether consumers, workers, and host communities embrace or resist the technology.