Sierra

Sierra is the clearest test yet of a simple wager: that an AI agent can actually finish customer service work, not just draft a reply for a human to check. Its co-founders picked the fight for a reason. Bret Taylor was co-CEO of Salesforce, CTO of Facebook, and a co-creator of Google Maps, and he now chairs the board of OpenAI. Clay Bavor spent 18 years at Google running Labs, AR/VR, and Workspace design.

They started the company in 2023 and moved fast. By late 2025 Sierra hit a reported $100M revenue run rate in under two years, and it now says roughly 40% of the Fortune 50 use its agents across chat, voice, email, SMS, WhatsApp, and ChatGPT. The company charges only when the agent delivers a result, which is either the boldest or the riskiest part of the whole plan.

Sierra builds and runs branded AI agents that talk to a company's customers and actually resolve things: return an order, process an insurance claim, answer a co-pay question, originate a mortgage. The pitch is that a single agent works across voice, chat, email, SMS, WhatsApp, and ChatGPT in 58 languages, always on, and connected to the systems of record it needs to take real action.

What separates it from a chatbot is the platform underneath. Sierra Agent OS lets companies build an agent, watch how it behaves, test it against edge cases, and keep improving it, with guardrails and a brand voice baked in so the thing stays on-message under pressure.

Contact centers have always been stuck choosing between quality and cost: hire more people for better service, or cut staff and watch wait times and frustration climb. Sierra's answer is an agent that handles high volume on its own, hands off to a human when it should, and keeps context across the whole conversation so customers do not have to repeat themselves.

The harder problem it takes on is trust. Enterprises will not point AI at billing, health benefits, or claims unless they can see why it did what it did, so Sierra leans hard on testing, observability, and the promise that customer data never trains its models.

This is an enterprise product, and Sierra does not pretend otherwise. Its named customers skew toward big, regulated, high-volume operations: health systems, insurers, telecom, banks, retailers, and travel brands. Reviewers make the flip side plain, noting the platform usually needs a dedicated engineering team to integrate and map complex flows, which leaves smaller businesses out.

The buyers are the CX and operations leaders drowning in call volume, plus the engineering teams asked to make an agent production-ready. Sierra sells to both, which is why it ships a no-code studio for one and an SDK for the other.

Sierra's most interesting idea is that you should not have to hand-build an AI agent at all. Ghostwriter takes your SOPs, transcripts, or a plain-English description and produces a working, multilingual, multichannel agent, then keeps testing and patching it. Explorer runs ChatGPT-style deep research across thousands of real conversations to tell you what to fix next, and can send that fix straight back to Ghostwriter.

Underneath sits Agent OS: an Agent SDK for engineers who want to write customer journeys as code, Agent Studio for teams who want no code at all, plus memory, decisioning, and observability so the agents get smarter with every conversation.

Sierra sells on outcome-based pricing, which it describes as charging only when its software achieves a specific, valuable result rather than per seat or per interaction. It is an aggressive way to sell software and a clean alignment story for a buyer worried about paying for an AI that does not work.

There is no public price list. Everything runs through enterprise sales and custom contracts, and reviewers repeatedly flag the pricing as opaque and expensive, which is the tradeoff for a model built entirely around large, negotiated deals.

Sierra is one of the loudest names in a crowded fight to run enterprise customer service with AI, up against conversational-AI platforms like Cognigy and a wave of agent startups, plus the incumbents whose seats it wants to replace. Its edges are the ones money and pedigree buy: a founder who ran Salesforce and sits on OpenAI's board, a full build-test-observe platform instead of a bare model, and outcome-based pricing that reframes the sale.

The strategic bet it keeps making is that the future is agents you describe rather than software you configure, and it is spending nearly a billion dollars of fresh capital to get there first.

The customer numbers are specific enough to be worth something. Minted reports case resolution above 65% and CSAT around 95% on its Sierra agent. Sutter Health, with 25 hospitals and over 3.5 million patients, uses agents inside its chronic disease program, and Sierra says its healthcare agents already reach more than half of U.S. families.

The logos back the scale story: ADT, Safelite, Clear, Casper, DIRECTV, SiriusXM, Ramp, SoFi, Vanguard, Rocket Mortgage, Wayfair, and more. These are the company's own selected wins, but they are named, dated, and tied to real metrics rather than vague claims.

The praise across reviews is consistent: Sierra chews through high chat volume, keeps the experience steady across channels, and its hands-on team makes integrations go faster than expected. Buyers who commit tend to like what they get.

The complaints are just as consistent. The platform has a real learning curve, it is not plug and play and usually demands a dedicated engineering team, pricing is opaque and steep, and some users report the agent can be slow to sense frustration and escalate. This is powerful software with a heavy lift, and the reviews reflect that honestly.

Sierra has raised roughly $1.585B, and the ramp is steep. In September 2025 it took $350M at a $10B valuation in a round led by Greenoaks. Eight months later, in May 2026, it raised about $950M at a $15.8B post-money valuation, this time led by Tiger Global and Google's GV, with Sequoia, Benchmark, and Greenoaks along for the ride.

The capital tracks the revenue: roughly $100M ARR in November 2025, then a reported $150M by early 2026. A $15.8B price tag is now the bar the company has to grow into.

Sierra has the money, the founders, and the logos to be a defining company in enterprise AI, and the ARR curve suggests demand is real. The open questions are the ones its own reviews raise: can it make deployments less heavy, its pricing less opaque, and its agents reliably good enough to justify a $15.8B valuation.

If outcome-based pricing holds up, meaning customers keep paying because the agents keep working, Sierra's bet pays off. If the results wobble at scale, that same pricing becomes the pressure it has to answer for.

Sierra is deliberately an in-person company, headquartered in San Francisco with offices in New York, Atlanta, London, Paris, Madrid, Munich, Singapore, Tokyo, and Sydney. It runs on five stated values, Trust, Customer Obsession, Craftsmanship, Intensity, and Family, and the careers copy leans into speed, taste, and getting the details right.

The culture reads high-intensity but not careless: leadership says balance and intensity can coexist and pitches Sierra as one of the best tech companies for parents. Everyone is expected to know the customers, and when one has a problem the team drops everything to fix it.

Sierra pays at the top of the market and discloses ranges across many roles and currencies. In the U.S., engineering bands run into the low $400Ks at the high end, with product, sales, and design in similar territory; it also posts bands in GBP, EUR, SGD, AUD, CAD, and JPY for its global offices.

These figures are base only. Full-time employees are eligible for Sierra's equity plans, and sales roles carry commission on top, so the real package for most people runs meaningfully higher than the salary line.

The legal entity behind the product is Sierra Technologies, Inc. Its core promise to enterprise buyers is that customer data is never used to train its models, and it treats the boundary between a live conversation and its systems as an engineering discipline: on payments, sensitive data enters as voice or chat and leaves as a token, so the agent never touches the plaintext.

Sierra also draws a clean line on data it processes for customers versus data it collects itself, and it says any request about that customer data goes to the customer, not to Sierra.

Sierra has become a fixture in AI business coverage, mostly for how fast its valuation and revenue are climbing. The through-line across the reporting is the same: a marquee founder, a hot category, and numbers that keep beating the last headline.

The recent trail runs from the $350M round at a $10B valuation in September 2025, through the $100M ARR milestone that November, to the roughly $950M raise at $15.8B in May 2026.