Bret Taylor’s Sierra just closed a $950 million round led by Tiger Global and GV, pushing its valuation past $15 billion and giving the company over $1 billion in war chest. The stated goal: become the default platform for AI-powered customer experiences. But the numbers behind the announcement tell a more interesting story about the operational reality of deploying agentic systems at scale.

Sierra claims over 40% of the Fortune 50 as customers, with agents processing billions of interactions—mortgage refinancing, insurance claims, returns, donor campaigns. That’s a massive throughput engineering challenge. The company hit $100M ARR in November, then $150M by February. That growth rate reflects both enterprise urgency and the steep compute costs of running production agents.

Uber CTO Praveen Neppalli Naga recently admitted his team “blew through our AI budget” after opening the door to agentic tools. The payoff? About 10% of code at Uber is now autonomously generated across 8,000 engineers. A hotel-booking integration that normally takes a year was completed in six months using agentic workflows. That’s a 2x improvement, but it came with an unplanned spike in infrastructure spend.

Sierra’s April launch of Ghostwriter—an “agent as a service” tool that takes natural language descriptions and deploys specialized agents autonomously—points to their architectural bet: abstracting agent creation into a meta-layer. Taylor’s thesis, laid out at HumanX, is that most enterprise software is barely used. Sierra’s play is to let users skip the UI entirely, routing intent directly to agentic workflows.

The engineering trade-off is clear: massive upfront infrastructure investment for potentially transformative efficiency gains. The question is whether the cost curve flattens before budgets do.

Source: TechCrunch