Lead Agentic AI Engineer

About the role

Engagement Type: Full-Time

About CodeRoad

CodeRoad is a technology execution partner trusted by companies across North America since 2002. We work alongside our clients to engineer, modernize, and scale the technology that powers their businesses, combining deep technical expertise, specialized teams, and shared accountability for outcomes. Our capabilities span product engineering, platform engineering, legacy modernization, and AI engineering.

Our engineers, designers, delivery leaders, and technology specialists collaborate through Velocity Studios: specialized teams developed within CodeRoad to deliver complex technology initiatives. Being assigned to a project means you become part of a collaborative environment where you can strengthen your expertise, take ownership of meaningful work, and grow alongside experienced professionals.

Why Build Your Career at CodeRoad?

-

High-Velocity Work & Client Exposure: We work close to the technology, teams and roadmaps that move real outcomes. Direct engagement with U.S. client teams and roadmaps.

-

Career Continuity & Long-Term Opportunity: Your career here can outlast any single project. Continuous multi-project pipeline; long-tenure CodeRoaders.

-

Engineering Autonomy & Technical Mastery: We invest in the skills that shape what's next. Agentic AI tools and training, Coursera Enterprise, Studio practice labs, L1-L7 Career Framework.

-

High-Trust Employee Experience: You grow in an environment built on trust, transparency and recognition. High-touch onboarding, transparent leadership communication, community recognition, psychological safety.

Key Responsibilities

-

Architect Multi-Agent Orchestration Systems: Design and deploy stateful multi-agent coordination graphs using LangGraph or custom DAG loops, enabling dynamic task decomposition, supervisor orchestration, human-in-the-loop escalation, and self-healing error recovery.

-

Engineer MCP Interfaces & Connectors: Build standardized Model Context Protocol (MCP) servers and connectors to securely expose Salesforce tools, custom objects, actions, and enterprise context to autonomous agent runners.

-

Optimize Reasoning, Latency & Token Spend: Architect multi-tiered LLM routing strategies across model tiers (utilizing Claude 3.5 Sonnet/Opus for complex multi-turn reasoning and Haiku for high-throughput utility tasks) to balance latency, accuracy, and API costs.

-

Establish Automated Evaluation Pipelines: Implement automated eval harnesses (Promptfoo, Ragas, DeepEval) within CI/CD pipelines to continuously measure context grounding, schema validity, and hallucination resistance against golden benchmark datasets.

-

Enforce Defense-in-Depth Security & Observability: Implement end-to-end telemetry and security using LangSmith or Langfuse, combined with input sanitization, field-level security (FLS) compliance, and automated PII scrubbing.