Sarah Wang's a16z thesis reframes AI's role in customer service from cost-reduction to experience transformation, arguing that when the marginal cost of attention approaches zero, every business can operate as a concierge rather than a scale operation. The argument hinges on a fundamental economic insight: whilst logistics scaled exponentially with the internet, human attention scaled only linearly, forcing companies like Delta and Verizon into a binary choice between low-cost, low-touch service or premium, high-touch concierge models reserved for luxury brands. AI collapses this constraint. Decagon's early results—80 percent deflection rates paired with doubled Net Promoter Scores at companies like Chime—demonstrate that automation need not trade quality for efficiency. Instead, AI agents with full customer context, always-on availability, and infinite parallel capacity can deliver proactive, personalised service at scale. This inverts the conventional narrative about AI replacing agents; it's not about headcount reduction but about unlocking latent demand for genuinely attentive customer relationships that were economically impossible before.
The implications for CX teams are substantial but require careful navigation. The concierge model Wang describes—where support dissolves into commerce, where AI anticipates problems before customers articulate them, where every interaction deepens rather than resets the relationship—demands a fundamental rethink of how support organisations measure success and structure their work. For teams already running conversational AI platforms, the question becomes whether your implementation is genuinely proactive and contextual or merely deflecting tickets more efficiently. The distinction matters because the former creates competitive moat through customer intimacy; the latter is commoditised. More pressingly, if this thesis holds, the primary interface between business and customer shifts from transactional support to continuous relationship management, which means support leaders must either expand their remit into commerce and retention or risk becoming peripheral to customer strategy. The risk is that organisations optimised for reactive ticket resolution—the current state of most enterprise support stacks—will struggle to operationalise the ambient, predictive engagement model Wang envisions, even with capable AI infrastructure in place.
The broader tension here is timing and capability. Wang's argument assumes AI agents will achieve the contextual sophistication and reliability required for proactive intervention at scale—knowing not just what customers bought but why they bought it, what they'll want next, and when they're about to churn. Current implementations, including Decagon's, excel at deflection and resolution but operate largely within reactive, problem-solving frames. Whether AI can genuinely shift to the ambient concierge model—where it notices a payment failure before the customer does, or recommends a product because it understands taste rather than just purchase history—remains an open question. For CX professionals, this creates an immediate strategic choice: invest in building the data infrastructure, customer intelligence, and proactive workflows now, betting that AI capability will catch up, or wait for clearer evidence that the concierge model is achievable beyond luxury segments. The former positions teams as architects of the future; the latter risks obsolescence if Wang's thesis proves correct.
The internet ruined customer service. AI could save it. a16z.com
The internet ruined customer service. AI could save it. Andreessen Horowitz