Natalie Meurer, Head of Agent Engineering at Sierra, spoke at the AI Engineer World’s Fair about how teams build conversational AI agents for enterprise customer service. She leads a global team of more than 120 engineers focused on designing and deploying these agents. Before Sierra, Meurer worked in technology policy, taught herself to code and spent five years at Palantir.
Forward deployed engineering (FDE) vs. agent engineer
Meurer explained that forward deployed engineering (FDE) started as a model for placing highly technical staff close to customers, but the label now covers a wide variety of roles across the AI industry. Sierra deliberately uses the title "agent engineer" to emphasize the technical shape of the work — building and integrating agents — rather than only the customer-facing aspect. According to Meurer, FDE today is defined more by accountability to customers than by a fixed set of skills.
What an agent engineer does
Sierra builds conversational AI agents for inbound and outbound customer service, including low-latency voice and chat agents as well as email-handling agents. The work combines data integration and systems engineering with what Meurer calls "taste": the ability to design what sounds human and reassuring in a voice agent.
Discovery and selecting use cases
Engagements often start with discovery rather than a predefined use case. Sierra looks for problems that are both technically challenging and have meaningful business impact. In financial services, for example, dispute processing is a typical starting point because it is complex and requires emotional intelligence: an agent may need to calm a customer who sees a fraudulent charge. Meurer says almost every Sierra customer is moving toward using an agent as a front-door IVR-style system — the first responder when a customer calls.
Much of the work sits above the model layer
Sierra treats agents as an orchestrated constellation of models. While the team continuously evaluates the best model for each task, most customer-specific effort happens at the orchestration layer and in integrations rather than inside the individual models. Sierra also integrates customers' own models at times and helps customers learn to build agents on the platform, applying internal knowledge and context.
Custom deployments and reusable patterns
Projects mix bespoke and reusable elements. Every agent is intentionally specific to its organization — reflecting brand, APIs, systems and operating procedures — but some capabilities, such as answering from a knowledge base, are broadly reusable. Sierra maintains industry expertise across financial services, healthcare, travel and hospitality, and retail. Meurer says large organizations have reached production in as little as 40 to 60 days, while still customizing agents to their environments.
Agents as enterprise systems: orchestration and change management
"Agent" can mean the phone-answering entity, a sub-agent, or even a prompt with tools. Enterprises need ways to maintain an agent ecosystem: manage integrations, coordinate teams, and follow discrete release processes. That's a change-management problem as much as a technical one. Sierra tends to model a single agent as owning an entire customer interaction, composed of subtasks they call "journeys," but enterprises still need governance so hundreds or thousands of contributors can work together safely.
Convergence of product engineering and FDE
Meurer believes product engineering and customer-facing engineering are converging. As code becomes cheaper to produce, customer insights can be translated more directly into product. The best product engineers should be talking to customers, and forward deployed engineers should be building product. Customer-facing accountability remains essential: even with very capable reasoning models, processes still need to be encoded to ensure consistent handling of repetitive customer-service situations.
Why generalists may gain value
Meurer expects the most valuable developers to develop product and customer-facing skills. While there are arguments that specialists will remain important, she leans toward generalists gaining relative value because FDE has traditionally required both engineering and customer-facing competencies. Agent engineers and forward deployed engineers occupy a forward-looking area in AI and engineering practice.
Titles and the future of engineering roles
The market currently uses multiple labels — go-to-market engineers, forward deployed engineers, agent engineers, AI engineers — and Meurer expects these to become different parts of the broader engineering craft. New engineering jobs will emerge as technology and organizational needs evolve.
Conclusion
For Meurer, agent engineering in enterprise customer service is a practical, customer-centered discipline combining integration, orchestration and product sensibility. While FDE roles are diverse, the defining feature today is accountability to customers and the ability to translate what teams learn into robust, maintainable enterprise systems.
