Pol Peiffer’s core message was that companies should “rent the intelligence and own the context.” Speaking from Sierra’s work with enterprise AI agents, he argued that access to frontier models will not be a durable differentiator because everyone will be able to use the same models and the next generations of them. He also pushed the point further by saying that the data companies already have today is not necessarily the advantage either. Instead, the long-term advantage will come from the context built as AI agents interact with customers, learn from those interactions, and compound those learnings inside the company rather than at competitors.

He framed the talk around three recurring CIO concerns he hears across San Francisco, London, and Munich: how to differentiate when everyone has the same intelligence, how to avoid ceding value to frontier-model labs, and how to produce ROI from AI investments. Peiffer predicted that over the next one or two years, almost all customer conversations will be mediated by agents. Those agents will become the front door of businesses, interacting with customers, reasoning, acting, and taking action on their behalf. As agent conversations become cheaper, companies will have more of them, enabling deeper one-to-one customer relationships and more personalized experiences.

Peiffer emphasized that building production-grade agents is much more complex than combining a model, tools, a knowledge base, and retrieval augmented generation. Beneath the “iceberg,” agents need to handle languages, accents, phone, email, chat, WhatsApp, empathy, assertiveness, and scale across millions of conversations, including regulated industries like banking and healthcare where mistakes are unacceptable. He then showed how agents can redefine customer experience: Redfin lets users describe home preferences in natural language and renders the interface around that intent, while Wilson Sports uses an agent to help customers navigate a large product catalog and surface signals that can improve product pages and eventually inform product development.

He closed the main talk with examples of agents evolving from one-time support-ticket handlers into systems that own customer relationships over weeks and months. Figr, a U.S. home equity lender, uses an agent by phone and text to help customers through document uploads, credit pulls, identity checks, explanations, and tax information before handing off to a loan officer. Peiffer said loans handled by the agent plus loan officer showed a 143% higher close rate than loan officer alone, emphasizing top-line revenue rather than just cost reduction. In Q&A, he explained that Sierra works closely with customers, productizes best practices, uses Ghostwriter to help build agents, and sees better results when customers train their own teams to maintain agents tied to core business workflows.