AI Engineer in Cybersecuirty

AI Engineer

**Global Enterprise Security AI Engineering**

Team Overview

GES AI Engineering is a small, high-leverage team that builds agentic AI and automation capability for Global Enterprise Security as a whole — not for a single program. The team is chartered to build once, reuse everywhere: shared agent platforms, evaluation tooling, and automation patterns that any GES team can adopt.

The team is four AI Engineers that are assigned to GES priorities as they arise rather than to one fixed team’s tooling. An engineer’s specific assignment will change over time as GES leadership reprioritizes; what stays constant is the underlying skill set below. Engineers cross-cover each other’s assignments during surges, incidents, or absences.

Role Overview

The AI Engineer designs, builds, and operates agentic AI workflows and automation on behalf of Global Enterprise Security’s teams — most often as an embedded partner rather than the workflow’s eventual owner. This is a builder role: the engineer scopes a partner team’s problem, prototypes an agent or automation against it on a shared platform (for example AWS Bedrock or an equivalent managed model platform), proves it is accurate and safe, and either hands it to the partner team to operate or keeps it running as a shared GES capability.

Assignments are expected to vary — a given engineer might spend a quarter embedded with a detection team, the next building a fleet-scale automation, and the one after that prototyping something for a team that has never used agentic tooling before. Because of that, this position is hired against durable capability, not against a fixed backlog: how quickly and safely the engineer can turn an unfamiliar security workflow into a working, trustworthy automation.

What You’ll Do

Problem Scoping and Solution Design

• Take an ambiguous, partner-team-described problem in any security domain and scope a narrow, testable first version of it — clarifying what “good” looks like before writing code.
• Choose the right automation or agentic pattern for the problem at hand (deterministic automation, retrieval-augmented response, tool/function-calling agent, multi-step orchestration) rather than defaulting to the most complex option available.
• Identify which parts of a partner team’s workflow are high-volume and low judgment (good automation candidates) versus which require human decision-making and design the human/agent boundary accordingly.
• Assess data availability, access, and integration requirements early, and flag when a proposed assignment isn’t yet ready for an AI-assisted solution.

Agentic Workflow and Automation Engineering

• Build agents and automation on the organization’s shared model and orchestration platform(s) (for example AWS Bedrock or a comparable managed platform), including model selection/routing, tool and function definitions, and context and retrieval management.
• Build integrations and tool interfaces — including MCP servers or equivalent connectors — that give an agent scoped, least-privilege access to the data and systems it needs, and no more.
• Develop and maintain automation or agent logic as code: version control, peer review, staged promotion from development through production, and rollback capability.
• Write clean, maintainable Python and API integrations that a teammate can pick up and extend without the original author in the room.

Evaluation, Safety, and Production Hardening(Following skills are plus but not mandatory)

• Build an evaluation harness for every assignment before calling it done — golden examples, precision/recall or task-success measurement, and hallucination/false-action detection — and use it to decide whether a prototype is ready to progress.
• Define human-in-the-loop approval gates for any destructive or high-impact action, and ensure every automated or agent-driven action is logged, attributable, and reversible where feasible.
• Monitor deployed workflows for drift, degraded accuracy, and cost or rate-limit issues, and respond before they become incidents.
• Apply a bias toward measured autonomy: earn expanded authority for automation by demonstrating accuracy in the partner team’s environment, not by assumption.

Shared Platform Contribution and Reuse

• Contribute to the shared agentic AI platform used across GES — model routing, guardrails, connectors, and evaluation tooling — so the next assignment starts from a stronger foundation than th…