Sierra
Enterprise agent platform for chat, voice and email with a declarative Agent SDK, GitHub Actions CI/CD and simulation testing. Outcome-based pricing with no published rate card.
Overview
Founded in 2023 by Bret Taylor and Clay Bavor, Sierra sells customer-facing agents to large enterprises. The named customer list runs to Rocket Mortgage, Vanguard, SiriusXM, Uber, Nubank, Ramp, Hyatt and Wayfair, and the company claims over 40 percent of the Fortune 50. A $950M raise in 2026 valued it at $15.8 billion on roughly $200M ARR. Channels cover chat, voice, email, SMS, WhatsApp and ChatGPT.
What separates it from the rest of the category is the Agent SDK. Agent behaviour is written in a declarative language rather than assembled in a drag-and-drop builder, skills compose, and changes flow through GitHub Actions like any other repository. Simulation suites run before deploy. Traces expose every tool call and every branch the agent took. Sierra also published τ-bench and τ²-bench, the tool-agent-user benchmarks a lot of the industry now cites, plus τ-voice covering 278 grounded speech tasks across retail, airline and telecom. Publishing your own evaluation harness openly is a reasonable proxy for confidence.
Horizon, announced 16 July 2026, is the larger bet: a single goal that persists across many conversations over days or weeks, such as originating a loan or chasing a prior authorization. Genuinely useful if your work spans sessions. The weakness is commercial rather than technical. There is no pricing page, no entry tier and no sandbox, so getting the SDK in front of your own data means an enterprise procurement cycle first. A team that wants a working agent this month will get there faster on Fin or Decagon and can revisit Sierra when the volume justifies the process.
Key Features
- ✓ Agent SDK with a declarative language for goals, guardrails and composable skills rather than a click-together builder
- ✓ CI/CD through GitHub Actions so agent changes move through normal code review and deploy paths
- ✓ Simulation and test suites that run before release, plus trace inspection down to individual tool calls
- ✓ Horizon, launched July 2026, carries one goal across many conversations over days or weeks
- ✓ Low-latency voice stack with streaming ASR and turn-taking, in production at retail and financial services customers
- ✓ Channels span chat, voice, email, SMS, WhatsApp and ChatGPT from shared agent definitions
Where it holds
- • Treats agent behaviour as source code with tests and CI, which almost nobody else in this category does
- • Full trace inspection makes a production incident debuggable instead of guesswork
- • Publishes τ-bench, τ²-bench and τ-voice openly on GitHub, so its evaluation claims can be independently rerun
- • Long-horizon goals in Horizon fit workflows like loan origination or prior authorization that genuinely span sessions
Where it breaks
- • No pricing page, no tiers and no trial, so comparison against Fin or Decagon is guesswork until you are deep in procurement
- • Enterprise-only motion means a small platform team cannot prototype on it this quarter
- • The widely quoted $1.50 per resolution comes from third-party trackers, not from Sierra
- • Outcome definitions shift under Horizon, and a multi-week goal is a harder billing unit to audit than a resolved chat
My Take
The interesting part for an engineering lead is not the agent, it is the SDK. Behaviour is declarative source, skills compose, changes ship through GitHub Actions and a trace replays down to the individual tool call, which is the difference between debugging an incident and guessing at it. Sierra also open-sourced τ-bench and τ²-bench, so the benchmark claims can be rerun rather than taken on faith. Then the commercial reality lands: no pricing page, no trial, no self-serve tier and a reported $1.50 per resolved conversation that comes from outside trackers rather than Sierra itself.
Quick Info
- Pricing:
- paid
- Openness:
- Proprietary
- Starting at:
- No published pricing of any kind. Sierra sells enterprise contracts priced per achieved outcome, and third-party trackers report roughly $1.50 per resolved conversation, a figure Sierra has never confirmed. No tiers, no self-serve plan and no free sandbox, so evaluation runs through a scoped procurement cycle. Since the July 2026 Horizon launch an outcome can be a multi-week business goal rather than a single resolved conversation, which changes the unit you are billed on. Company raised $950M in 2026 at a $15.8B valuation on reported ARR around $200M. Checked sierra.ai, August 2026.
- Added:
- Aug 2026
- Updated:
- Aug 2026
Use Cases
Judge it on your own work
The notes above say where Sierra holds and where it breaks. The fastest check is your own workload.
Visit website ↗Alternatives to Sierra
Fin
paidAI support agent that runs on top of Zendesk, Salesforce, HubSpot or Intercom and bills $0.99 per outcome. Renamed from Intercom in May 2026, now under a $3.6B Salesforce acquisition.
Coze / 扣子
freeByteDance’s AI bot-building platform that enables anyone to create custom AI assistants with plugins, workflows, and knowledge bases without coding.
Writer
paidEnterprise AI content and agent platform on Writer's own Palmyra models, with a knowledge graph, brand guardrails and audit trails. No free tier, Starter is $29 per user a month annually.