Natalie Meurer

Natalie Meurer founded Sierra’s agent engineering function and helped build customer-facing AI agents, including a Sonos agent for troubleshooting speaker connections. In July 2026, Latent Space identified her as Sierra’s Head of Agent Engineering, leading a global team of more than 120 engineers. She also publishes as Nat Meurer.[1][5][7]

Meurer and her colleagues intentionally chose agent engineer rather than forward deployed engineer to describe the technical work more specifically. In her July 2026 interview, she argued that forward deployed engineering is defined more clearly by accountability to customers than by a consistent skill set. Sierra’s role combines systems integration and agent development with understanding customer operations and the end-user experience; designing voice agents also requires judgment about what sounds natural.[1]

Her Sonos work illustrates that combination. In “Meet the AI agent engineer”, published July 11, 2024, she described helping build an agent to accelerate “time to music” by troubleshooting speaker connection issues. She and technical counterparts first scoped the problem in detail, including the Sonos product portfolio, then translated that understanding into an agent intended to solve the problem repeatedly for different customers.[5]

Meurer distinguishes answering questions from completing transactions. Retrieval can supply a return policy, but returning a purchase or changing a subscription requires access to APIs and systems of record. Production workflows often require multiple API calls in a particular order. In her 2024 account, she explained how Sierra’s team-developed Agent SDK composes skills and enforces deterministic API interactions so agents reach the appropriate systems at the appropriate point in a workflow.[5]

Her account describes two complementary controls: supervisor models that check agent behavior, and explicit limits on where generative output is appropriate. Where regulatory requirements demand precise wording, the SDK can selectively bypass an LLM. She also described her team’s work building reusable components from common agent behaviors, then extending them for particular deployments. These abstractions allow an agent’s capabilities to change without rebuilding its behavior around a single specialized model.[5]

By July 2026, Meurer characterized most customer-specific work as occurring at the orchestration layer rather than within models. Her team’s discovery process seeks difficult problems with meaningful business impact. She used financial-services dispute processing as an example: an agent must follow the process correctly while responding sensitively to someone frightened by a fraudulent charge. She also emphasized the organizational work of maintaining integrations, coordinating contributing teams, and following a discrete release process.[1]

Meurer argues that cheaper code authoring makes it easier to translate customer insights directly into products, bringing product engineering and forward deployed engineering closer together. As she put it in the Latent Space interview: “If you are a product engineer, you should be talking to customers. If you are a forward deployed engineer, you should be building the product.” She leans toward the view that generalists will become more valuable as engineering incorporates more product and customer-facing work.[1]

That argument rests partly on the kind of problem customer-service agents solve. Meurer contrasts coding agents tackling new problems with service agents handling variations of the same process repeatedly. Even a highly capable reasoning model, she argues, should follow an appropriately encoded return process rather than independently reconstruct it each time. Customer-facing engineers remain important because they help turn an organization’s rules and operational knowledge into consistent agent behavior.[1]

In November 2025, Meurer and Rosie Roca coauthored an account of Sierra University, describing agent building as a collaborative craft involving customer-experience, operations, and engineering teams. The three-day program moves from defining agent behavior to API integration, SDK control flow and state management, then building and stress-testing participants’ own use cases. Their shared rationale was direct: “We designed Sierra University because building great agents isn’t a solo act. It’s a craft.”[6]

Before establishing Sierra’s agent engineering function, Meurer worked across policy and software engineering. She holds a Georgetown University degree in Science, Technology, and International Affairs and graduated from Stanford Graduate School of Business. Recalling her earlier career in July 2026, she said she began in technology policy, learned to code on the side, and became an engineer on Palantir’s privacy team. She spent about five years there across law enforcement, defense, and infrastructure engineering, then attended business school to add a business dimension to her experience. She also previously founded a company.[4][1][5]

In a July 12, 2024 post, Meurer recalled weighing go-to-market, engineering, policy, and product opportunities after Stanford before choosing to build AI agents as one of Sierra’s first agent engineers. This earlier experience provides context for her description of the role as combining engineering expertise, customer understanding, and business curiosity.[7][5]