Written by Jesse Zhang
The following is a memo written by Decagon's CEO, Jesse Zhang, to all employees.
In the near future, every brand will have an AI agent that serves as its primary interface with customers. Need to book a flight? Report a lost card? Order new running shoes? The way every business interacts with its customers is fundamentally changing.
Every major technology shift like this produces a handful of defining companies. In the previous major shift, cloud replaced on-premise servers and unlocked a new category of applications. The successful ones, like Databricks, became $100B+ businesses. The shift to AI is even more significant, because it's touching a larger category of spend than cloud ever did. For the first time, delivering personalized, concierge-level experiences to every customer isn't constrained by headcount.
That’s the market Decagon is building for. We’re in the early innings of creating the next Databricks of this generation.
The application layer is where value accrues. Just as OpenAI and Anthropic are building the critical model layer, someone has to build the critical application layer on top that actually solves business problems end-to-end. History is clear on how this plays out: your hotel minifridge earns higher margins for a pack of M&M’s than the cocoa bean refinery. We have direct exposure to the business problem, with the pricing power that comes with it.
The scope is massive. Decagon's main use case is customer service. As we’ve grown, our product has evolved to handle all conversations a brand has with its customers i.e., the "AI concierge". The reason we're winning against incumbents and competitors is that we’ve built the best end-to-end product. The entire development lifecycle can be done right in the product, accessible to more than just engineers.
This is a genuinely hard technical problem. Strong foundational LLMs are only a small part of the solution.
The rest of the industry is taking a Palantir-style approach, where forward-deployed engineers assemble custom agents for each company. It sounds reasonable until you realize that AI agents need constant iteration, and that architecture makes iteration expensive and slow.
What businesses actually need is a natural language interface and an execution engine capable of capturing arbitrarily complex logic across wildly different customers. That requires training custom models to meet real performance requirements. It also requires building the surrounding platform tooling — versioning, analytics, simulation environments, automated self-improvement — so that less-technical teams can build and optimize their agent without engineering support.
To accomplish this, we’ve assembled high-agency teams of cracked engineers, applied AI researchers, and former/future founders.
The opportunity is in front of us, and it’s ours to execute on.
