The role
From ServiceNow's own posting.
The Forward Deployed Engineer, Voice AI Agent Engineering designs, builds, ships, and operates voice AI agents whose core behavior is model-driven rather than explicitly authored — conversational and agentic experiences that interpret caller intent over a live audio channel, reason over context, invoke tools, and act on the caller's behalf. The engineer works across the full lifecycle: shaping ambiguous call flows and telephony requirements into precise specifications, authoring the instructions and prompts that govern voice agent behavior, proving through automated evaluation that the result behaves correctly on real calls. This is not a machine learning or AI research role — the engineer does not train foundation or speech models — but it is distinct from traditional software engineering, where systems follow deterministic call-flow logic rather than selecting execution paths at runtime. 
Two things follow, and voice makes both harder. First, the agent's most important logic often lives in natural language — instructions, prompts, tool descriptions, and guardrails — which must be engineered, versioned, and reviewed with the same discipline as code, while also holding up under what a spoken channel adds: turn-taking, barge-in, and disambiguating what speech recognition got wrong. Second, because behavior is probabilistic and happens live on a call, correctness is established by measuring behavior at scale — including latency, containment, and transcription accuracy — rather than by asserting fixed outputs, making automated evaluation a first-class engineering activity rather than a quality-assurance afterthought. The engineer owns the correctness of what ships to a live caller, whether a human or an agent produced it. 
What You Get to Do in This Role: 
Build voice AI agents. Design and ship agentic voice applications — intent interpretation, multi-step reasoning, tool invocation, and action on the caller's behalf — together with the CTI, CCaaS, and telephony integrations that make them usable in production. 
Design AI-driven autonomous voice workflows. Decompose call flows and business processes into the steps and decision points a voice agent can execute — determining where autonomy is appropriate, where a live-agent handoff is required, and how misrecognition, retries, and escalation are handled. 
Engineer for what voice demands. Build experiences that hold context across turns, manage barge-in and interruption, recover gracefully from speech-recognition error, disambiguate caller intent, and confirm before consequential actions — all within the latency budget a live call imposes. 
Author and maintain voice agent instructions. Write, structure, and version the system instructions, role definitions, tool descriptions, guardrails, and escalation rules that govern voice agent behavior — as engineering artifacts under review and regression coverage, not configuration text. 
Engineer prompts for reliability rather than demo quality. Iterate prompts against measured call outcomes: task decomposition, golden call transcripts, structured output schemas, grounding and citation, graceful hand-back to a human agent, and the token and latency cost that directly shapes caller experience. 
Build automated evaluation and test non-deterministic voice behavior. Design and operate the evaluation that makes change safe on a live channel — golden call datasets, multi-turn conversation suites, model-as-judge scoring calibrated to human QA review, CI gates, and drift detection — plus adversarial, jailbreak, grounding, and tool-selection testing specific to voice. 
Specify precisely and direct AI coding agents. Convert customer call-flow and telephony requirements into testable specifications with explicit scope, constraints, and acceptance criteria; decompose work into agent-sized tasks; and review agent output for correctness and maintainability. You own the result regardless of what produced it. 
Deliver as a forward deployed engineer, embedded in the voice stack. Work directly inside a customer's telephony, CTI, and CCaaS environment — building against their call data, integrations, and channels, tuning instructions and evaluation sets against live traffic, and returning with evidence that improves the product. 
Own quality, safety, and reliability on live calls. Monitor call containment, hallucination, unsafe actions, and transcription accuracy; defend against prompt injection and data leakage on live calls; and feed production call failures back into specifications and evaluation sets. 
Collaborate across product, design, and engineering. Partner with product managers, designers and platform teams to define success criteria, align on tradeoffs, and communicate capability and risk clearly
To Be Successful in This Role:
A demonstrated track record of building, shipping, and operating production voice applications, including hands-on delivery of AI-native voice agent features that real callers depend on rather than demos. Direct experience authoring agentic instructions and prompts for voice, designing AI-driven autonomous call workflows, and building the evaluation and testing that verifies them against live call traffic, is required. Willingness to work directly with customers in a forward deployed capacity, embedded in their voice and contact center environment, is required. At least 10 years of experience in software engineering with at least 2 years of experience in deploying contemporary AI solutions at customers.  
Education:Bachelor's degree in Computer Science, Software Engineering, or a related technical field, or equivalent practical experience. Advanced degrees or certifications are a plus but are not a substitute for a demonstrated record of shipping reliable, production voice AI agents. 
Key Competencies, Skills, and Experience 
Software engineering fundamentals. Strong command of data structures, algorithms, system design, APIs, data modeling, and testing. 
AI native voice application development. Hands-on experience building applications where model-driven behavior is central — agent orchestration at the application layer, tool and function calling, context assembly, grounding against enterprise data, and handling the latency, cost, and failure modes unique to a live audio channel. 
AI-driven autonomous workflow design. Demonstrated experience designing workflows in which voice agents carry out multi-step call processes with limited supervision: call-flow decomposition, decision points and autonomy boundaries, human-in-the-loop checkpoints, exception and retry handling, and observability over what the agent did and why. 
Voice and telephony platform expertise. Practical experience with telephony integration, speech recognition and synthesis, and streaming audio — including the barge-in, turn-taking, and confirmation patterns a live call requires. 
Agentic instruction authoring. Demonstrated skill writing and maintaining the natural-language logic that governs voice agent behavior — instructions, role definitions, tool descriptions, guardrails, and refusal and escalation rules — with versioning, review, and regression coverage applied as they would be to code. 
Prompt engineering. Intent-driven prompt design tuned for voice: decomposition, golden and few-shot examples, structured output and schema enforcement, grounding and citation, and disciplined iteration against call-level evaluation results rather than impressions. 
Evaluation and testing of non-deterministic voice systems. Ability to build measurable frameworks assessing call quality, agent behavior, tool-selection accuracy, and regression risk — golden call datasets, scenario suites, model-as-judge scoring with human QA calibration, continuous evaluation pipelines, and drift detection — plus adversarial, jailbreak, and grounding testing. 
Specification precision and architectural