Decagon's challenge to the forward-deployed engineer model strikes at a fundamental tension in enterprise AI adoption. The role has exploded—job listings surged over 800% between January and September 2025—with Microsoft committing $2.5 billion and 6,000 embedded specialists, and AWS pledging $1 billion to similar programmes. Yet Decagon's CEO Jesse Zhang argues this proliferation signals product failure, not market necessity. His evidence comes from a single customer who built three workflows in a year with Sierra's embedded engineers, then built seven within a month after switching to Decagon. The implication is stark: if vendors require permanent technical staff to operate their systems, they've built tools too complex for the organisations buying them. This reframes the entire conversation around implementation costs—what looks like a service advantage (dedicated engineers) may actually be a dependency tax.
The tension between Decagon's thesis and Salesforce's scale reveals two competing bets on the same market. Salesforce's Agentforce reached $1.2 billion in annualized recurring revenue in Q1 2027, up 205% year-over-year, and the company acquired Fin for $3.6 billion. These figures reflect distribution advantage rather than direct product comparison, but they demonstrate that embedded implementation models can sustain massive revenue growth. What matters for CX teams already running Agentforce or considering Sierra is whether the engineering overhead represents temporary friction during adoption or permanent architectural necessity. Decagon's counter-argument—that roughly 90% of its workloads now run on fine-tuned open-source models rather than frontier models, delivering faster and cheaper performance on narrow tasks—suggests the real differentiation lies in whether a platform enables your own staff to iterate independently or locks you into vendor-controlled customisation cycles.
The stakes extend beyond vendor selection to budget allocation and team structure. If enterprise AI can achieve genuine self-service configuration at scale, the economics shift dramatically: vendors grow revenue without proportional increases in implementation headcount, and your organisation gains autonomy over workflow changes rather than waiting for reassigned engineers. If it cannot, forward-deployed engineers remain a permanent line item, and the question becomes whether you're paying for software or paying for the people required to make that software functional. For support leaders evaluating platforms, this distinction should drive procurement decisions—not the presence of embedded engineers, but whether the product architecture eventually empowers your team to own the system or perpetually requires vendor intermediaries.
Enterprise AI’s Hottest Job Just Found Its Biggest Skeptic PYMNTS.com