The usual explanation for Sierra is founder gravity: Bret Taylor ran Salesforce, chairs the OpenAI board, and co-created Google Maps. That explains the first meeting. It does not explain the fastest enterprise ARR curve of this cycle.
The actual mechanism
Six choices that reinforce each other. Sierra prices the outcome: a pre-negotiated rate per resolved case, escalations to humans free, so its revenue depends on the product working. It owns the last mile with forward-deployed agent engineers instead of throwing software over the wall, then productizes what they learn into Agent Studio and Ghostwriter. It built an engineering discipline for non-deterministic software, where customer-experience staff annotate real conversations daily and every annotated failure becomes a permanent regression test. It builds capabilities ahead of the frontier models, like best-in-market Cantonese voice, and throws the code away without sentiment when models absorb it. It lands one high-volume channel at the biggest brand that will say yes, proves resolution and CSAT, then expands to voice plus chat. And it markets with credibility artifacts, the τ-bench and τ-voice benchmarks and published engineering doctrine, instead of category slogans.
The full playbook breakdown covers each with the how-to-steal-it version.
The one-sentence version
Every choice is the same decision made six times: put yourself on the hook for the outcome, on purpose. Sierra is what a software company looks like when its P&L, its engineers, its release process, and its go-to-market are all accountable for whether the product lands rather than whether it ships. Most vendors are structurally insulated from their own failures. Sierra deleted the insulation, and the market is paying for the difference.