AI Agent Engineer, Cloud Infrastructure

Orbis is building the agentic AI capability that supports its mission operations. The AI Agent Engineer designs and operates multi-agent workflows and owns the cloud infrastructure, security, networking, and cost management those workflows depend on.

Responsibilities:

• Design, build, and deploy agentic AI workflows. Architect and ship the multi-agent and automated workflows that run in production: triage, data retrieval, summarization, reporting, and escalation routing
• Translate in both directions. Explain to non-technical stakeholders what the systems do, where they are reliable, and where they are not, in terms those stakeholders can act on. Equally, take a process owner's plain-language description of how their work happens and turn it into a workflow specification that holds up
• Partner with process owners to build what they actually need. Work directly and iteratively with the colleagues who own a process so the resulting capability fits their operation rather than forcing them to fit it. Push back constructively when a stated requirement will not survive contact with production
• Keep the workflows reliable. Monitor what is running in production, diagnose failures across prompts, tools, data, and orchestration, and resolve them. Know which problems are yours to solve and when to bring in engineering
• Own the cost of execution. Track what your AI workloads cost to run. Identify where a tighter prompt, a caching layer, or a simple deterministic step delivers the same outcome, and keep those costs defensible as volume grows
• Set the quality bar for AI-assisted output. Define accuracy standards, human review checkpoints, and escalation criteria for the automated workflows, and build the checks that prove those standards are being met
• Document what you build. Author and maintain the runbooks and SOPs that allow others to operate and troubleshoot these workflows without you. Nothing lives only in one person's head
• Integrate across Orbis products. Work across Catalyst, Pulse, and Discovery, with engineering support where needed, so workflows can access and act on the data they require
• Deliver the reporting leadership and the team depend on. Build and maintain the outputs behind resolution rate, escalation rate, output accuracy, response time, and employee satisfaction, and turn the Senior Operations Manager's requirements into structured, recurring reports and data exports
• Close capability gaps and keep the practice current. Recognize when a need is not served by existing tooling, scope the gap clearly, and either build it or bring it to engineering and product with enough detail to act on. Track developments in agentic AI tooling and enterprise automation platforms, and form a defensible view on what is worth adopting, at what pace, and for which use cases
• Design, deploy, and maintain the Orbis cloud footprint across GCP, AWS, Cloudflare, and Azure, spanning both enterprise workloads and client delivery environments
• Define and manage that footprint as infrastructure as code so environments are reproducible, reviewable, and recoverable
• Own networking, identity, and secrets handling for agent workloads, including the boundaries between environments and the audit of access
• Hold the security posture of the platform and keep it evidenced rather than asserted
• Track and control platform run cost across compute, inference, storage, and egress, and report it in terms leadership can act on
• Build and maintain the deployment and release path — CI/CD, environment promotion, rollback — so shipping an agent workflow is routine rather than an event

Requirements:

• Must have 5 years of proven experience translating user requirements into wireframes, workflows and such
• Must have 1-year proven experience utilizing Claude or similar commercial offering to build AI agents
• Demonstrated experience building a functioning multi-agent workflow, with the ability to walk through the design rationale, how the agents hand off work and share context, what broke, and how it was fixed
• Practical command of prompt engineering, including retrieval-augmented generation (RAG) and tool or function calling
• Proven ability to explain technical work to non-technical audiences — describing a system's behavior, limits, and failure modes to a process owner or executive without hedging or retreating into jargon
• Demonstrated ability to elicit requirements from stakeholders who cannot hand over a specification, and to reach a working solution regardless
• Ability to troubleshoot agent workflows built on platforms such as Claude, GPT-class models, LangChain, AutoGen, or CrewAI, and to isolate whether a failure originates in the prompt, the tool, the data, or the orchestration
• Working proficiency with APIs and sufficient scripting ability to build and debug workflows independently; software engineering experience is not required, but a JSON payload or a Python script must not be a blocker
• Sound judgment about model output, including the discipline to state plainly when an output should not be trusted
• Strong written communication. This role produces documentation, runbooks, and reports that program leads and executives rely on
• Hands-on production experience with at least one major cloud platform (GCP, Azure, or Cloudflare) and working familiarity with a second
• Practical experience managing infrastructure as code using Terraform, Pulumi, or an equivalent
• Working command of cloud networking, identity, and secrets management, and the ability to treat security posture and run cost as design constraints rather than afterthoughts
• 5-7 years experience in software development
• Experience coding using Python
• Kubernetes or serverless runtimes at production scale; multi-cloud or cross-boundary deployment; cloud migration or consolidation work
• FinOps or formal cloud cost management practice; zero-trust architectures; federated access control models
• owning CI/CD pipelines end to end; infrastructure monitoring, logging, and alerting for production workloads
• prior work as a solutions consultant, technical program manager, forward-deployed engineer, or in a product-adjacent role
• monitoring, evaluation, or observability tooling for AI systems in production; automated evaluations or regression checks for LLM-based workflows; human-in-the-loop review steps that maintain quality without creating bottlenecks
• designing SOPs or escalation frameworks that hold up across multiple stakeholders
• working fluency with Claude, Copilot, or Cursor
• operating within or supporting a government or defense program; OSINT, DIGINT, or intelligence analysis workflows and tradecraft; data governance concepts, federated access control models, or zero-trust architectures
• a track record of coming up to speed quickly on unfamiliar systems, and effectiveness in a distributed, multi-program environment with competing priorities