The proliferation of text-based AI agents—from Instinct's $10 billion valuation to Caddy's iMessage integration to Fambot's family-focused approach—represents venture capital's latest consensus bet. But beneath the convenience narrative lies a troubling assumption: that users will comfortably cede meaningful control to systems that operate with limited oversight. This market is growing precisely because investors believe consumers will grant these agents sweeping permissions. The question is whether they should.
Consider what's actually happening. Instinct now gives users 'dedicated email addresses for their assistants' so agents can sign up for services, contact businesses, and handle tasks 'that require an email account.' Folk 'runs code and executes multi-step tasks' across devices. Martin can 'send messages or make phone calls' on users' behalf. Comma maintains 'context and memory across devices and conversations.' These aren't convenience features—they're delegation of agency itself. The report acknowledges that Instinct's autonomous email capability 'has also raised concerns about privacy and security,' yet the company proceeds anyway. This is the venture model in action: move fast, raise capital, and let regulators catch up.
What's particularly revealing is how these agents compete on the premise of reducing friction. Ask Proxy costs $1 per day for personal use but $15 to $23 monthly per employee for businesses—pricing that suggests the real customers are companies willing to pay to monitor and automate their workforce. Fambot launches free during beta expecting to eventually 'charge roughly the equivalent of a Netflix subscription,' banking on the sunk-cost fallacy of family coordination. Folk offers free beta with $8.33/month Pro, and Martin charges $21 monthly. The business model depends on making delegation irresistible, then monetizing the dependency.
The fundamental problem is that these platforms are optimizing for seamlessness rather than accountability. Haven's 'digital twin' concept—understanding 'what you would do and how you would do it'—sounds helpful until you consider what happens when the system makes an error, commits to something on your behalf, or misinterprets your preferences. Who bears responsibility then? The agent? The company? You? The report doesn't address this, and neither do most venture investors, who are betting that scale precedes regulation.
The market for AI agents is real and may become essential infrastructure. But the current trajectory—where convenience drives adoption before safeguards exist—is irresponsible. Before Instinct reaches ubiquity and before Fambot becomes the default family organizer, the industry needs enforceable standards: explicit audit trails for agent actions, genuine user consent mechanisms (not buried terms of service), and clear liability frameworks. These aren't obstacles to progress. They're the prerequisites for trust.