Sierra AI Pays Up to $390,000 for Engineers Who Never Write a Button — and 132 Open Roles Show the Enterprise UI Is Already Dead

The $15B Bet on Agentic Enterprise Software

Sierra AI closed a $950 million Series E in May 2026 at a $15 billion post-money valuation, led by Tiger Global and GV, Google's venture arm, according to TechCrunch. Benchmark, Sequoia, and Greenoaks participated. Eight months earlier, in September 2025, the company had raised $350 million at $10 billion, TechCrunch reported. In that window, Sierra added roughly $50 million in annualized recurring revenue, going from $100 million ARR to $150 million, and pushed past 40% Fortune 50 penetration.

The numbers are aggressive. At $15 billion against $150 million ARR, the implied multiple is 100x. That's rich even by current AI benchmarks, but the trajectory underwrites it: Sierra crossed $100 million ARR within seven months of its commercial launch in February 2024, a ramp the company says is among the fastest in enterprise software history.

The company was founded in early 2024 by Bret Taylor and Clay Bavor. Taylor spent nearly a decade at Salesforce, co-founded Quip (acquired by Salesforce for $750 million in 2016), and now chairs OpenAI's board. He held that position through the November 2023 board crisis that nearly cost Sam Altman his job. Bavor spent close to twenty years at Google leading products including Google Workspace, Google VR, and Google Labs. The pairing is deliberate: Taylor brings enterprise CRM distribution instincts and AI-strategy credibility; Bavor brings experience shipping AI-first products at scale.

That founder pedigree matters more than it would in most markets. Enterprise procurement is slow, deliberate, and trust-dependent. Security reviews, compliance assessments, and executive sponsorship all gate large deals. Sierra's Fortune 50 customer list, assembled in under two years, suggests the founders' reputations shortened sales cycles in ways a less credentialed team could not have replicated.

Sierra's platform builds conversational AI agents for customer-facing workflows: mortgage refinancing, insurance claims, product returns, nonprofit fundraising. These are not chatbots layered onto existing systems. They are purpose-built agents that handle full transaction lifecycles autonomously. Sierra says its agents have processed billions of customer interactions across hundreds of millions of end users, with customers including SoFi, Ramp, and Brex.

In April 2026, the company launched Ghostwriter, a product that signals where the strategy is headed. Ghostwriter lets users describe a task in natural language ("an agent that handles renewal conversations for business insurance customers") and autonomously creates and deploys a specialized agent configured for that task. The target is to turn AI agent deployment from an engineering project requiring specialist talent into something closer to a product decision.

The $950 million raise is earmarked for geographic expansion into Europe and Asia and for extending the platform beyond customer support into sales and customer lifetime value optimization. The international push is timed to a window where enterprise AI agent adoption outside North America is still early. The sales expansion is the harder bet: customer service is a cost center with a clear ROI story; sales is a revenue center where agents have to prove they can drive incremental revenue, not just cut headcount.

Taylor's continued role as OpenAI board chair gives Sierra an unusual vantage point on frontier model development. It's not a formal partnership, but the adjacency informs the company's thinking about what agentic AI can do and how fast it will get there. Whether that translates into a durable competitive moat — as foundational models commoditize and competitors like Salesforce's Agentforce and ServiceNow's Now Assist chase the same budgets — is the open question hanging over the $15 billion price tag.

Inside Ghostwriter: Building Agents Without Writing Code

Sierra AI launched Ghostwriter in late March 2026 with a blunt premise: building AI agents for customer experience should not require editing journeys, writing integrations, or creating simulations by hand. Bret Taylor, Sierra's co-founder and CEO, framed it as a shift from "clicks to prompts": describe the behavior you want in plain English, and Ghostwriter produces a production-ready agent that spans voice, chat, email, and more than 30 languages.

The platform operates in three phases. In the Build stage, a user uploads SOPs, raw support transcripts, whiteboard photos, or audio recordings of subject matter experts (or simply types a description of the desired behavior). Ghostwriter ingests that material, extracts key behaviors and edge cases, and generates a complete agent with guardrails, tone, and style configured through prompts rather than code. Every change is visible for review before anything goes live, so teams can audit what Ghostwriter built before deploying it.

The Test phase runs automatically. Ghostwriter generates simulations with every build or update, stress-testing the agent against edge cases beyond the obvious scenarios. When a simulation fails, the platform diagnoses the issue and implements the fix itself, with no manual debugging loop. Sierra's product materials describe this as closing the development loop without engineer intervention.

In the Improve stage, a companion tool called Explorer analyzes live customer conversations and surfaces recommendations ("build empathetic support for cancellations," for instance, or "add secure subscription management"). Each recommendation carries a one-click "Fix with Ghostwriter" button. Teams can also feed the system recordings of their top-performing support associates, and Ghostwriter uses those to refine agent behavior over time.

Taylor, announcing the release on LinkedIn, drew a direct line to the coding world: "Codex and Claude Code have transformed how we build software, making it possible for software engineers to orchestrate and review the work rather than doing all the work themselves. We think the same transformation will happen for all software." His claim is that every enterprise platform's UI will eventually become an agent that does the work on a user's behalf, rather than a web app a human clicks through.

That thesis has hiring implications. If the interface shifts from screens and forms to natural-language prompts and autonomous agents, the engineering work shifts with it. The focus moves away from frontend feature development and toward the systems that let agents act safely inside real business operations. Sierra's own open roles already reflect that: a Product Manager, Agent Development and a Forward Deployed Infrastructure Engineer sit alongside the more traditional positions on its growing team.

132 Open Roles: What Sierra's Hiring Tells Us

Sierra AI's careers page lists 132 open roles. Five of those went up in the past week alone, according to Zero G Talent's board data, a pace that signals the company is still deep in expansion mode despite already operating offices across three continents.

The heaviest concentration is in Agent Engineering, the discipline Sierra essentially named. The company has posted 21 distinct "Software Engineer, Agent" listings spanning San Francisco, New York, Atlanta, London, Singapore, Madrid, Paris, Tokyo, Munich, Toronto, and Sydney. Several of these are language-specific roles (Arabic, Cantonese, Dutch, French, German, Italian, Korean, Spanish, and Thai) a pattern that maps directly onto Sierra's enterprise customer base. If you're building agents for a European telecom or an Asian retailer, the engineer who configures that agent needs to speak the customer's language, not just Python.

Compensation details for key roles:

These are not entry-level numbers, and the job descriptions make clear why.

What Sierra Actually Wants

The Agent Engineer listing lays out a workflow that looks nothing like a typical SWE role. You own the full Agent Development Life Cycle (pilot, deploy, iterate) and you work directly with enterprise customers to scope the problem before writing a single prompt. The example projects tell the story: building an agent for a telecom that outperforms human agents at reducing subscription churn, troubleshooting Sonos speaker connections, navigating furniture delivery logistics through apartment door codes.

The required skills split into two tiers. The baseline: production systems experience, comfort with ambiguity, customer-facing communication, and what Sierra calls "high agency" — a builder's instinct to ship around obstacles. The "even better" list is where the AI-specific bar sits: experience deploying LLMs in production, familiarity with eval frameworks, RAG pipelines, prompt engineering, and agent tooling. React, TypeScript, and Go are called out as preferred languages. Founding experience is a plus, not a requirement; Sierra says it values people who've balanced craft, ownership, and speed.

The blog post on the agent engineer role adds technical depth. Engineers at Sierra work with frontier models (GPT-4o, Claude 3.5 Sonnet, Gemini), vector databases, and orchestration frameworks like LangChain or Sierra's own Agent SDK. The SDK is central to the workflow: it lets engineers compose "skills" in a declarative language and stack them into agents, rather than hard-coding behavior into a fine-tuned model. The post describes a supervisor architecture where a second model verifies the first agent's output, pushing combined accuracy from 90% to 99%. That's the kind of system design the role demands.

Where the Jobs Are

San Francisco is the hub. The majority of platform engineering, infrastructure, security, and product roles list SF as the sole or primary location. New York and Atlanta serve as secondary US hubs, particularly for agent engineering and sales. London is the European center, with agent engineering, product, and GTM operations roles. Singapore, Tokyo, Paris, Madrid, Munich, Toronto, and Sydney each have a smaller but targeted set of openings — usually agent engineers and product managers matched to regional language needs.

Sierra's careers site says the company is "primarily an in-person company" and that interviews happen onsite. Remote roles exist but are the exception: the US Remote designation appears on a handful of sales and sales engineering positions, and Canada has a remote option. If you're applying for an engineering role, plan to be in an office.

Beyond Agent Engineering

Sierra is also hiring across Platform Engineering (18 listed roles), Product (15+ roles across agent development, data platform, Ghostwriter, voice, and infrastructure), Sales Engineering (6 roles), and Developer Relations (3 roles). The platform roles (infrastructure, SRE, security, payments, identity) are almost entirely San Francisco-based and look like standard high-scale platform hiring. The agent-specific roles are where the geographic spread and language specialization get unusual.

Zero G Talent's board currently lists 5 roles added in the past week, including a People Partner for Europe in London and a Forward Deployed Infrastructure Engineer in London. That's a slower clip than the 38 roles OpenAI added in the same window, but Sierra is a younger company hiring into a more specialized niche. The roles that are going up suggest the company is still building out its European operations and tightening its enterprise deployment infrastructure — the unglamorous plumbing that has to work before agents can handle millions of customer conversations.

The Enterprise UI Is Disappearing — Here's Why

Gartner predicts that by 2028, 33% of enterprise software applications will include agentic AI, up from less than 1% today. Deloitte's 2025 Emerging Technology Trends study found that while 38% of organizations are piloting agentic solutions, only 11% are running them in production. The gap between experimentation and deployment is where most of the hiring, and most of the engineering difficulty, lives.

What's actually changing

Traditional enterprise UIs were built around a constraint: humans are slow, error-prone, and need hand-holding. So software exposed a fraction of its capability at a time, through carefully designed screens that guided users step by step. Every button was a decision the designer made about what you should be allowed to do next.

Agentic systems invert that. Microsoft's Richard Riley, general manager for agents and low code at Power Platform, frames it directly: "Instead of teaching people how to use systems, we can let people express intent — and allow systems to determine how that intent is carried out." The agent becomes the interaction surface. The user says "resolve this case" or "prepare a customer briefing," and the agent orchestrates the steps across systems, policies, and data sources.

Scott Havird, who writes about enterprise architecture, argues this goes deeper than a chatbot layer on top of existing apps. He describes an emerging architecture where AI agents sit directly between users and data layers, eliminating not just the UI but the application servers that traditionally sat in between. "Apps no longer need to expose every possible action through UI," he writes. "Instead, they provide trusted capabilities the agent can invoke, enforce business rules and permissions, and act as systems of record, not systems of navigation."

That's a fundamentally different job for the software. The app is no longer where work happens. It's how work is made possible.

The talent implication

When the UI disappears, the engineering challenge shifts from layout and interaction design to orchestration, context management, and multi-agent coordination. You need people who can build systems that understand business intent, route it across services, and handle the failure modes when an agent misinterprets a request or hits a policy boundary.

The Deloitte report notes that leading organizations are discovering agentic AI requires "building agent-compatible architectures, implementing robust orchestration frameworks, and developing new management approaches for digital workers." That's a different skill set than CRUD app development. It's closer to distributed systems engineering, with a layer of natural language understanding and business logic reasoning on top.

Where the UI still matters — and where it doesn't

Not every interface is vanishing. Salesforce reports that 87% of U.S. consumers experience frustration with traditional customer service interactions that require multiple steps and complex menus. The agentic replacement (conversational, context-aware, proactive) is a direct response to that friction.

But the shift is uneven. Toyota's supply chain team now uses an agent that delivers real-time vehicle shipment information without anyone interacting with the mainframe — a process that previously required 50 to 100 mainframe screens. HPE built an agent called Alfred that conducts internal operational performance reviews by coordinating four underlying agents for data analysis, visualization, and report generation. These are end-to-end process replacements, not UI tweaks.

At Mapfre Insurance, AI agents handle routine claims administration tasks like damage assessments, but customer communication still goes through a human. Maribel Solanas Gonzalez, Mapfre's group chief data officer, calls it "hybrid by design." The agent handles what it can safely and efficiently do. Everything else stays with people.

The governance problem hiding inside the UI problem

Here's what makes this hard: removing the UI removes the guardrails. Traditional interfaces constrained what users could do by only showing them certain buttons. When an agent can invoke any capability in the backend, the constraint has to come from somewhere else — policy engines, permission systems, audit logs, human-in-the-loop checkpoints.

Gartner predicts over 40% of agentic AI projects will fail by 2027 because legacy systems can't support modern AI execution demands. The systems lack real-time execution capability, modern APIs, modular architectures, and secure identity management. The UI was doing more work than anyone realized — not just presenting options, but enforcing boundaries by omission.

Moderna took this seriously enough to combine its technology and HR functions under a single chief people and digital technology officer. Tracey Franklin, who holds that role, says the logic is straightforward: "The HR organization does workforce planning really well, and the IT function does technology planning really well. We need to think about work planning, regardless of if it's a person or a technology."

What engineers should take from this

The button-clicking era produced a generation of engineers who specialized in translating business requirements into screens. The agentic era needs engineers who can translate business intent into orchestrated workflows — systems that reason about what needs to happen, coordinate across services, and fail gracefully when context is ambiguous.

That's a harder problem. It's also why Sierra AI, OpenAI, and the rest are hiring aggressively for it. The UI isn't disappearing because designers got bored. It's disappearing because the interface was never the point — the work was. And agents, when they work, let you skip the interface and go straight to the outcome.

Sierra vs. Salesforce vs. OpenAI: Three Bets, One Market

Sierra AI's hiring push doesn't exist in a vacuum. It's running headlong into two well-funded competitors building agentic tools for the same enterprise buyers, and the differences in approach reveal how fragmented this market still is.

Salesforce has the most established position. At Dreamforce 2025, the company announced general availability of Agentforce 360, a platform that wraps AI agents into every layer of its CRM stack. The pitch is integration: agents grounded in Salesforce's Data 360 layer, embedded in Customer 360 apps, and surfaced through Slack as what Salesforce calls the "Agentic OS for the Enterprise." The company says more than 12,000 customers are already using Agentforce, with results like Reddit deflecting 46% of support cases and cutting resolution times by 84%, and OpenTable resolving 70% of diner inquiries autonomously.

Salesforce's strategy leans on its existing enterprise footprint. It's not asking companies to rip out their CRM — it's layering agents on top of the data and workflows customers already have. The expanded partnerships with OpenAI (GPT-5 integration in Agentforce 360, ChatGPT commerce) and Anthropic (Claude models for regulated industries like financial services and healthcare) give Salesforce customers model choice while keeping everything inside Salesforce's trust boundary. Slack's rebuilt AI assistant, now in pilot with 70,000 Salesforce employees, adds a conversational layer that pulls from workspace context, files, and connected calendars.

OpenAI is attacking from the other direction. Its Frontier platform, announced in 2025, is an enterprise agent framework with shared context, permissions, and governance — aimed at companies that want to build agents directly on OpenAI's models rather than through a CRM intermediary. OpenAI's partnership with Salesforce shows the overlap: GPT-5 is available as a reasoning engine inside Agentforce 360, and ChatGPT users can access Salesforce data and Tableau visualizations through natural conversation. But OpenAI is also building its own enterprise agent tools, competing for the same engineering talent. Zero G Talent's board lists 38 OpenAI roles added in the past week alone, spanning research engineering, forward-deployed engineering, and business operations.

Sierra AI sits between these two. It doesn't have Salesforce's 26-year CRM installed base or OpenAI's frontier model research. What it has is a $15 billion valuation and Ghostwriter, a platform purpose-built for customer-facing agent engineering (not a general-purpose CRM add-on or a model API wrapper). Sierra's five new roles on Zero G Talent's board in the past week are concentrated in agent development and infrastructure deployment, suggesting the company is scaling its platform team rather than building out a broad horizontal suite.

The hiring patterns reflect these different bets. Salesforce is cutting roles in some areas — SF Ben reported layoffs across Agentforce, Marketing Cloud, and MuleSoft even as the platform launches — while OpenAI is aggressively expanding headcount across research and go-to-market. Sierra's smaller, more focused hiring blitz suggests a company still in build mode, trying to own a specific slice of the agentic stack before the bigger players fully converge on it.

The question for engineers evaluating these companies isn't who has the best agent technology — it's whose platform will actually get deployed at scale. Salesforce has the customers. OpenAI has the models. Sierra has the focus. The hiring numbers suggest all three believe the window to define enterprise agentic software is still open, but not for long.

Skills That Pay: What Engineers Need to Pivot Into Agentic AI

Zero G Talent's board shows Sierra AI added 5 roles in the past week alone, including a Forward Deployed Infrastructure Engineer in London at £170,000–£290,000 and two Product Manager, Agent Development roles in San Francisco at $180,000–$390,000. These are not research positions. They are production-facing roles that require shipping agentic systems to real customers, which means the hiring bar is built around deployment fluency, not model architecture papers.

The skill set that keeps appearing across Sierra's open roles and the broader agentic engineering market breaks into three layers.

Layer one: the non-negotiable foundation. Python proficiency, solid API design, and comfort with cloud infrastructure (AWS, GCP, or Azure) are table stakes. Every AI engineering job posting in 2025 requires them, and Sierra is no exception. If you cannot write clean, testable Python and deploy a model behind a REST endpoint, nothing else on this list matters yet.

Layer two: the agent-specific toolkit. This is where the market is paying a premium. Retrieval-augmented generation (RAG), prompt engineering at production scale, orchestration frameworks like LangGraph and CrewAI, and evaluation/observability for agent behavior. The Johns Hopkins Certificate Program in Agentic AI (an 18-week online program with live faculty sessions) covers exactly this stack: RAG, multi-agent systems, AgentOps monitoring, and security for autonomous agents. It is one of several programs responding to surging demand, alongside offerings from NVIDIA (its Agentic AI LLMs Professional Certification) and IBM's RAG and Agentic AI Professional Certificate on Coursera. These programs exist because the skills are new enough that most working engineers did not encounter them in school.

Layer three: the skills that separate candidates. Evaluation and observability for agent systems — knowing how to measure whether an agent is actually completing tasks correctly, not just generating plausible text. Security thinking around prompt injection, tool misuse, and bounded autonomy. And the operational discipline to containerize, monitor, and roll back agent deployments the same way you would any other production service. Addy Osmani's open-source Agent Skills repository on GitHub (64.6k stars) packages exactly these practices into structured workflows that production teams follow: spec-driven development, test-driven implementation, security hardening, observability instrumentation, and staged rollouts.

The practical takeaway: you do not need a PhD to enter this field. Only 27.7% of AI engineering job postings require one. You need a portfolio of deployed systems that demonstrate you can build, monitor, and fix agentic software running with real data. Start with a free resource — Anthropic's prompt engineering guide, DeepLearning.AI's short courses on agents and RAG, or LangChain Academy — then build a multi-agent system that solves a concrete problem and put it on GitHub. That portfolio will do more for your candidacy than any certification alone.

The engineers who will thrive in Sierra's orbit are the ones who treat agentic AI not as a chatbot wrapper but as a production engineering discipline with its own testing, monitoring, and security requirements. That bar is rising fast.

Working in AI? Zero G Talent tracks the openings: browse AI jobs, openings at OpenAI, Sierra Space and Sierra AI, and the people building the field.