customer-experience-ai · 201-1000 people · snapshot 2026-07-16 · not company-verified · sierra.ai
Documented stack
Category
Tool
Since
Confidence
Source
language
GoNamed directly as Sierra's backend systems language across independent job postings (frontend, payments infrastructure) and confirmed in production use for Sierra's internal agent-runner infrastructure.
PythonConfirmed for Sierra Research's open-source evaluation tooling (tau-bench family, all Python repos). Not directly confirmed as a production agent-platform language — only listed as one acceptable option among Go/Scala/Python in a data-platform job posting.
AWSExplicitly named as Sierra's cloud provider by the SRE team's own posting (distinct from a separate, weaker 'AWS, GCP, or Azure' generic candidate-skills list found in another posting, which was not used as evidence).
KubernetesConfirmed running Sierra's internal 'Pinecone' agent-runner platform. Not separately confirmed for the customer-facing production agent runtime, though plausible given shared infrastructure culture. Sierra states the same infrastructure also underpins Ghostwriter, its customer-facing "agent for building agents" — so this extends beyond internal tooling.
DatadogNamed as one of several acceptable observability tools (alongside Prometheus, Grafana, OpenTelemetry) in two independent job postings, always as an 'e.g.' example rather than an explicit 'we use X' statement. Included at inferred confidence only because the same short list recurs independently across two job families.
RedisRedis Streams used for live communication within Sierra's internal agent-runner platform (Pinecone). Not separately confirmed for the core customer-facing product.
Custom multi-model orchestration ("Constellation of Models")Sierra deliberately avoids single-vendor lock-in: agents are assembled from 15+ frontier, open-weight, and proprietary models, routed per task (latency-sensitive tool calling, classification, long-context reasoning, tone) by a custom orchestration layer. Sierra's own communications do not name the specific model vendors behind this system for the production platform.
Custom LLM-as-judge evaluation ("Monitors")In-house always-on evaluation layer; each monitor goes through its own evaluation loop grounded in team-labeled conversations and multi-model agreement.
LiteLLMConfirmed via Sierra Research's own open-source code as the model-provider abstraction layer for the tau-bench evaluation tooling. Not confirmed for the production agent-serving path.
OpenAIOpenAI's Codex is confirmed in company-wide internal use as a coding agent, and Sierra's own tau2-bench voice benchmark integrates OpenAI's realtime API as a supported provider. Sierra has not publicly confirmed OpenAI as a model provider behind its customer-facing agent platform — the company deliberately describes a vendor-agnostic 'constellation of models' without naming production vendors.
AnthropicClaude Code confirmed in company-wide internal engineering use. Not confirmed as a named production-platform model vendor — same caveat as the OpenAI entry. The Pinecone post also references Anthropic model names in default internal routing patterns.
Methodology: research assisted by AI agents, verified and edited by a human. Spot an error? Suggest a correction via pull request or email us. Work at Sierra? Verify your record and get the ✓ badge.