AI Engineer

Overview As an AI Engineer specializing in Agentic AI enablement, you will participate in the design and delivery of production-grade agent capabilities built on the enterprise AI Backbone across cloud and edge environments – across supply-chain and global functions. You will be responsible for end-to-end delivery of key agent modules and integration patterns (MCP/tooling), establish strong evaluation and regression discipline, and drive adoption by partnering with transformation teams, BU, platform engineering, and enterprise application owners. You serve as a technical engine for the workstream—translating business workflows into measurable agent outcomes, working to mitigate identified risks, evaluating/experimenting with options/tradeoffs, and working to scale solutions across domains.

Responsibilities Agent Engineering & Workstream Delivery (35%)

• Lead design and productionization of high-leverage agent modules and reusable patterns (tool-use orchestration, policies/guardrails, memory, RAG where it adds measurable value), built as composable components and reference implementations. *(Execute/Lead)*
• Translate ambiguous product/problem statements into concrete agent behaviors and system designs: state models, failure modes, tool contracts, latency budgets, and acceptance criteria that engineering + product can execute against. *(Execute/Consult)*
• Deliver quickly without sacrificing quality: create thin vertical slices, iterate with evidence, and converge on robust behavior under real-world constraints. *(Execute)*
• Drive meaningful performance gains via systematic optimization: latency, token efficiency, tool-call success, retrieval quality, and cost per successful task, including remediation of long-tail failure modes. *(Execute)*
• Proactively identify platformizable opportunities: refactor one-off implementations into shared frameworks/SDKs that reduce build time for others. *(Execute/Influence)*

Evaluation, Testing & Release Quality (25%)

• Define and implement evaluation strategies for assigned workflows: golden sets, scenario coverage maps, regression suites, online/offline metrics, and release gating thresholds aligned to real business outcomes. *(Execute/Consult)*
• Build repeatable evaluation systems (templates, labeling guidance, dataset/versioning conventions, dashboards/reports) so evaluation becomes a productized capability, not ad hoc testing. *(Execute/Lead)*
• Implement robust automated testing across layers: unit tests for prompt/tool wrappers, contract tests for tool schemas, integration tests for toolchains, and agent simulation tests for multi-step flows. *(Execute)*
• Lead root-cause analysis of quality failures (hallucinations, tool misuse, retrieval misses, routing errors): isolate causes (prompt/tool/data/model), implement corrective actions, and prevent regressions. *(Execute)*
• Champion evidence-first iteration: decisions and releases are backed by eval results, not gut feel. *(Influence)*

Model/Prompt Routing Contributions (15%)

• Contribute to router design and task-to-model mapping through routing rules/classifiers, prompt strategies, and model selection policies; validate decisions using evaluation data and runtime telemetry. *(Execute/Consult)*
• Propose and implement routing improvements when constraints change (pricing, latency, throughput, new model capabilities), with governance-aware rollouts and rollback plans. *(Consult/Execute)*
• Identify and mitigate routing failure modes (over-escalation to expensive models, under-routing causing quality loss, brittle heuristics) and improve robustness using lightweight ML or rules where appropriate. *(Execute)*

Integration with Tools and MCPs (15%)

• Lead implementation of MCP connectors/clients for enterprise apps and internal data products with strong engineering hygiene: schema/versioning discipline, typed contracts, scopes/permissions, auditability, and integration test strategy. *(Execute/Consult)*
• Build reusable integration patterns: standardized tool metadata, error normalization, retries/timeouts, idempotency, pagination handling, and consistent auth patterns to accelerate onboarding of new tools. *(Execute)*
• Collaborate with security/data owners to ensure secure-by-design tool access (least privilege, logging, PII handling, policy enforcement). *(Consult/Execute)*

Operational Readiness, Collaboration & Continuous Improvement (10%)

…