What is Sierra?

Sierra is an enterprise AI agent platform for customer experience. Rather than a chatbot bolted onto a helpdesk, it’s pitched as “Agent OS” — a system for building, deploying and continuously improving autonomous agents that handle the full customer interaction across every channel and, crucially, take real actions: processing returns, updating subscriptions, saving cancellations, ordering replacement cards, even authenticating patients or helping customers apply for a mortgage by connecting to back-end systems via API. It does not replace your ticketing system; it sits in front of it as the agent layer, escalating to humans when needed.

The company was founded in 2023 by Bret Taylor — co-creator of Google Maps, former co-CEO of Salesforce and current chairman of the OpenAI board — and Clay Bavor, an 18-year Google veteran who led Labs, Lens and AR/VR. That pedigree, plus a clear bet that AI agents become the “digital front door” for every brand, has made Sierra the company most enterprise buyers benchmark the category against.

The pedigree and the momentum

Few software companies have scaled this fast. Sierra hit $100M ARR in roughly seven quarters, crossed $150M+ by early 2026, and raised a $950M Series E in May 2026 at a $15.8 billion valuation (led by Tiger Global and GV), giving it over $1 billion to spend. Its customer roster spans tech-native and traditional enterprises alike — Deliveroo, Discord, Ramp, Rivian, SoFi, Brex and Tubi alongside ADT, Vans, Cigna, WeightWatchers and SiriusXM — with half of customers above $1B in revenue. It has opened offices from Tokyo and Singapore to Madrid, Paris, London and Sydney. This matters for a buyer because category momentum compounds: more deployments mean more model tuning, more reference customers and a faster product roadmap. The flip side is that Bret Taylor’s dual role as Sierra CEO and OpenAI board chairman remains an unresolved governance question, especially as OpenAI pushes further into enterprise deployment.

How Sierra works

Under the hood, Sierra uses a “constellation” architecture that orchestrates multiple large language models — OpenAI for some tasks, Anthropic or Google for others — to balance accuracy, latency and cost, rather than betting on a single model. Its Agent Data Platform gives agents memory, so a returning customer isn’t treated as a blank slate and context carries across channels. And in March 2026 it launched Ghostwriter, an “agent that builds agents”: you upload SOPs, transcripts, recordings or a plain-English description and it generates production-ready agents across voice, chat and email in 30+ languages. Alongside Agent Studio 2.0, a no-code/low-code builder, this is Sierra’s answer to its biggest historical criticism — that changing an agent’s behaviour required going through Sierra’s team.

Key features

Voice and omnichannel — the standout

Sierra runs one agent across voice, chat, email, SMS and WhatsApp, and voice has become its primary channel — it surpassed text in 2025 and the platform now handles hundreds of millions of calls a year. Given that the majority of customer-service contact still happens by phone, mature, low-latency voice that can actually resolve calls (not just deflect them) is Sierra’s single biggest advantage over chat-first rivals.

Action-taking resolution

Sierra’s agents don’t just answer — they act. Connected to your back-end systems, they complete returns, subscription changes, cancellations, account updates and transactions end to end, which is the difference between a deflection tool and one that genuinely resolves. Taylor has been explicit that the ambition reaches beyond support into sales and retention — agents that handle upgrades, identify cross-sell and run save-the-cancellation conversations at scale.

Agent building and continuous improvement

Ghostwriter and Agent Studio 2.0 lower the barrier to creating and iterating on agents, while Sierra’s Explorer tooling drives continuous, autonomous improvement from real conversations. The model is increasingly “build with or without engineering,” though in practice the largest deployments still lean on Sierra’s services team.

Scorecard

Overall score: 7.8 / 10 — the average of the eight categories above.

Pricing

Sierra has no pricing page — no tiers, no self-serve plans, no published per-resolution rate. Pricing is negotiated per deployment based on volume, complexity and integration scope, and you commit before you can model ROI independently. The figures below come from third-party reporting and should be treated as estimates; confirm directly with Sierra.

The honest read: outcome-based pricing is Sierra’s best idea and its biggest source of fine print. Paying when the agent actually resolves — rather than per seat or per message regardless of result — is genuinely customer-aligned, and it’s a major reason Sierra wins enterprise deals. But the model lives or dies in the contract details: what exactly counts as a “successful outcome”? If a customer comes back about the same issue two days later, is that a new billable resolution? Are routing and greeting interactions billed separately? Who pays when the AI handles most of a ticket and a human finishes it? Can you audit the resolution data independently, or does Sierra grade its own homework? Add ongoing optimisation and tuning fees, a likely annual minimum, and the cost of keeping a separate helpdesk, and the true total runs well past the headline rate. Negotiate the outcome definition, reporting and audit rights as hard as the price.

Pros & cons

Honest weaknesses

The first barrier is accessibility. Sierra publishes nothing on price, offers no self-serve trial and sells exclusively through a high-touch enterprise motion, so you’re committing six figures and a multi-week implementation before you can independently prove the ROI. That suits a Fortune 500 CX team with a procurement function and a transformation budget; it actively excludes the vast majority of support teams, for whom a helpdesk-native agent delivers most of the value at a tiny fraction of the cost and effort. Sierra is also explicitly a strategic-partner model rather than plug-and-go — powerful when you want a co-built, brand-perfect agent, frustrating when you need to change a policy within hours.

The second is the fine print around its headline strength. Outcome-based pricing is excellent in principle, but “resolution” is a slippery unit, and vendors that define and measure their own outcomes hold the pen on your invoice — so reporting transparency and audit rights matter as much as the rate. There’s also genuine structural risk worth naming: Sierra’s agents depend on third-party frontier models, so changes to model pricing, access or reliability flow through to Sierra; and the chairman-of-OpenAI question hangs over the company as OpenAI itself moves into enterprise deployment. None of this dents the product’s quality — Sierra is the best in its class — but it shapes who should actually buy it, and on what terms.

Who is Sierra for?

Sierra is the right pick for large enterprises that want the best autonomous agents — especially for voice — and can absorb the cost and the rollout: high-volume, phone-heavy operations in retail, financial services, telecom, travel and healthcare that see CX as a strategic, revenue-driving function rather than a cost centre, and that value outcome-aligned pricing. If that’s you, Sierra is the benchmark. It’s the wrong tool for small and mid-size teams, anyone who needs transparent pricing or a free trial, or teams that want to own and iterate their agent fully in-house without a vendor partnership. If you want the strongest agent inside a modern helpdesk with published per-resolution pricing and a free trial, Intercom Fin is far easier to adopt; for a comparable managed enterprise agent, Ada; and for the complete enterprise suite, Zendesk AI.

FAQ

AI customer supportSierraconversational AIAI agententerprise AIAI tool review