AI Platform Engineer

Forterra; Arlington (VA); On-site; USD 125,000 - 145,000; Full time

Benefits offered by this job

Premium Healthcare

Life/AD&D

Paid PTO

Parental leave

Tuition reimbursement

401(k) with match

Holiday calendar

Beverage fridge

Job summary

Forterra seeks a senior engineer to design and own a substrate for agent-based systems. You will define agent definitions, MCP servers, and guardrails to ensure trustworthy outcomes across a broad engineering landscape.

You will collaborate across systems engineering, safety, software, product, IT, cybersecurity, and leadership, with direct visibility to executive leadership. This role emphasizes architecture, governance, and scalable agent infrastructure.

Qualifications

• 3+ years building software systems, with meaningful recent depth in LLM-based or agentic systems.
• Fluency with modern agent architecture: tool/function calling, MCP or equivalent, context management, retrieval, multi-agent orchestration, and policy enforcement hooks.
• Judgment about model selection and cost/latency/quality tradeoffs, and tracking frontier rather than last year assumptions.
• Experience evaluating AI system quality with rigor.
• Experience with analysis of unstructured data.
• Experience running non-interactive workloads in production: CI pipelines, scheduled jobs, or event-driven triggers.
• Strong Python intuition plus comfort in shell, CI, and Linux service operation.
• Experience building and maintaining production applications.
• Working understanding of enterprise identity and authorization (OAuth2/OIDC/SAML, tokens, scopes, least privilege).
• Excellent written AND oral communication

Responsibilities

• Design agentic systems, not prompts.
• Author agent definitions and context as engineered artifacts, with explicit role separation, tool allowlists, model selection, and hard behavioral limits.
• Design multi-agent architectures where roles are adversarial and reliable by construction.
• Build skills with disciplined triggering, progressive disclosure, and non-overlapping ownership boundaries.
• Orchestrate work across multiple model vendors, harnesses, and agent runtimes.
• Make and defend the call on whether a problem warrants an agent at all, and engineer away from an agent left polling in a loop.
• Make agents part of the engineering infrastructure.
• Define the autonomy gradient explicitly: which agents are permitted to act, and which may only propose.
• Engineer unattended agents for bounded scope, idempotent reruns, fail-open vs fail-closed decisions, cost/runtime caps.
• Design human-facing escalation when blocked, and how fleets stay observable to humans.
• Prove agent output is trustworthy.
• Design defensible evaluations of AI tooling: controlled trials, ablations, blinded judges, leakage controls, uncertainty reporting.

Skills

LLM-based systems

Agent architecture

Model selection

Production workloads

Python

OAuth/OIDC/SAML

Communication

Job description

About Forterra

At Forterra, we are unleashing autonomy at scale to transform the battlefield. Our mission is to build the foundational platforms that enable an intelligent ecosystem to coordinate, adapt, and execute with speed and precision even in the uncertainty and disruption of modern conflict. In an era marked by rapid technological change and evolving threats, we design for flexibility, survivability, and operational dominance.

Forterra delivers weapons, sensors, and battlefield effects through integrated autonomous networks reaching operational areas faster, safer, and without placing human lives at risk. Our systems operate with distributed control, dynamic routing, and real‑time responsiveness, enabling sustained advantage across complex mission environments.

About the role

Forterra runs its engineering on agentic systems, not chatbots. Requirements, safety analysis, test generation, code review, and validation are increasingly performed by fleets of agents operating against Forterra's real systems of record.

This role builds and owns that substrate: the agent definitions, skills, MCP servers, guardrails, and evaluation harnesses that determine whether hundreds of engineers get leverage or just burn tokens.

The hard part is not wiring an API. It is judgment, such as knowing which problems should be handed to an agent, which should stay deterministic, when to use a classical data analysis method, and how to prove an agent's output is trustworthy.

You will work across the entire company: systems engineering, safety, software, product, IT, cybersecurity, and executive level leadership. Almost none of it sits inside your reporting line. Influence and growth directly come from performance and earned credibility. You will have direct visibility to chief executive leadership.

What you'll do

• Design agentic systems, not prompts
• Author agent definitions and context as engineered artifacts, with explicit role separation, tool allowlists, model selection, and hard behavioral limits
• Design multi‑agent architectures where roles are adversarial and reliable by construction
• Build skills with disciplined triggering, progressive disclosure, and non‑overlapping ownership boundaries - so capability scales without the context window collapsing
• Orchestrate work across multiple model vendors, harnesses, and agent runtimes
• Make and defend the call on whether a problem warrants an agent at all. Know when a deterministic script or webhook is the correct answer, and how to engineer away from an agent left polling in a loop
• Make agents part of the engineering infrastructure
• Design how agents get invoked across the full range of trigger patterns — a developer working interactively at a terminal, a nightly scheduled run, an event‑driven trigger like CI, and long‑running unattended work. Know the failure modes for each, and ship the result as shared, versioned engineering infrastructure rather than one‑off scripts
• Define the autonomy gradient explicitly: which agents are permitted to act, and which may only propose
• Engineer unattended agents for the absence of a human: bounded scope, idempotent reruns, deliberate fail‑open versus fail‑closed decisions, cost and runtime caps, structured verdicts, and epistemic scoped credentials that keep them auditable
• Design the human‑facing side of autonomous work: how an agent escalates when it is genuinely blocked, how and when it asks for a human in the loop, and how a fleet of concurrent agent sessions stays observable to the humans accountable for them
• Prove agent output is trustworthy
• Design defensible evaluations of AI tooling: controlled A/B trials, ablation studies, blinded LLM judges, contamination and leakage controls, reported uncertainty, and honest results
• Build validation harnesses that distinguish genuine signal from hollow green. Instrument your own systems against complexity drift: measure whether each added skill, agent, or tool is actually pulling its weight in quality, cost, and turns
• Build the guardrails
• Design structural controls for agents operating near controlled information: layered redaction, egress filtering, enforcement at the data layer rather than the prompt layer
• Anticipate cross‑boundary failure modes unique to agents
• Connect agents to the enterprise
• Build and operate the connective tissue between agent tooling and Forterra's infrastructure & processes: requirements management, issue tracking, documentation, source control, diagramming, and unstructured data
• Own enterprise identity integration end‑to‑end, including federated and government‑cloud authentication, and specify precisely what changes IT needs to make
• Run these as real production services with real uptime expectations, including incident response and root‑cause write‑ups
• Drive adoption
• Turn tribal knowledge into onboarding paths, curricula, setup automation, and self‑diagnosing tooling that lets an engineer's own agent fix its broken configuration
• Teach transferable technique rather than button‑clicking, including curriculum for non‑software audiences — systems engineering, safety, program management, design, business operations — and design yourself out of delivery so training scales past you
• Maintain the platform other engineers contribute to: review their work, set and hold the standards for acceptable agent patterns, grow the contributor base, and convert user pain into prioritized organizational demand

Minimum Qualifications

• 3+ years building software systems, with meaningful recent depth in LLM‑based or agentic systems
• Demonstrated fluency with modern agent architecture: tool/function calling, MCP or an equivalent tool protocol, context management, retrieval, multi‑agent orchestration, and hooks or middleware for policy enforcement
• Practical judgment about model selection and cost/latency/quality tradeoffs, and a habit of tracking the frontier rather than relying on last year's assumptions
• Experience evaluating AI system quality with rigor
• Experience with analysis of unstructured data
• Experience running non‑interactive workloads in production: CI pipelines, scheduled jobs, or event‑driven triggers
• Strong Python intuition plus comfort in shell, CI, and Linux service operation
• Experience building and maintaining production applications
• Working understanding of enterprise identity and authorization (OAuth2/OIDC/SAML, tokens, scopes, least privilege)
• Excellent written AND oral communication

Preferred Qualifications

• Strong background in statistics and/or classic machine learning AND deep learning
• Experience fine tuning LLMs on codebases and enterprise data (DPO, QLoRA, GRPO)
• Contributions to open source agent harnesses
• Knowledge‑graph, ontology, or retrieval architecture work across heterogeneous unstructured enterprise data
• Experience developing metrics and measuring productivity of software engineering
• Experience running an internal developer platform or enablement function: adoption metrics, contributor onboarding, deprecation
• Background in systems engineering, safety engineering, or verification & validation
• Experience with AI systems in a regulated, classified, or export‑controlled environment — CUI, ITAR, NIST 800‑171, FedRAMP/GovCloud
• Experience designing human‑in‑the‑loop review gates for automated systems
• Published, blogged, spoken, or contributed open source on agent evaluation methodology
• Familiarity with robotics, autonomy, or safety‑critical software development

Salary and Benefits

US Salary Range$125,000 - $145,000

The salary range for this role is an estimate and is based on a wide variety of compensation factors. The salary offered to candidates will vary based on a variety of factors including (but not limited to) relevant work experience, education, specialized training, critical expertise, training, and more. Equity in Forterra is included in most of our full‑time, high‑demand roles and is therefore considered part of Forterra's overall compensation package. In addition to base salary and equity, Forterra offers competitive benefits for full‑time employees including:

• Premium Healthcare Benefits: Three plan options, including an HSA‑eligible plan, with Forterra covering 80% of the plan premium for you and your dependents.
• Basic Life/AD&D, short and long‑term disability insurance plans 100% covered by Forterra, plus the option to purchase additional life insurance for you and your dependents.
• Extremely generous company holiday calendar including a winter break in December.
• Competitive paid time off (PTO) offering 20 days accrued per year.
• A minimum of 7 weeks fully paid parental leave for birth/adoption.
• A $9k annual tuition reimbursement or professional development stipend.
• Fully stocked beverage refrigerators with all the Celsius your little heart desires.
• 401(k) retirement savings plan, including traditional, Roth 401(k), and after‑tax deferral with company match up to 4%.

CLEARANCE ELIGIBILITY - This position may require eligibility to obtain and maintain a U.S. security clearance. For more information about U.S. Security Clearances:click here.

Forterra is an equal‑opportunity employer, providing and promoting equal employment opportunity in all personnel practices. Forterrans are unique, talented individuals who are united through a shared passion to deliver autonomous systems that enable national resilience and a robust supply chain.

All qualified applicants will receive consideration for employment without regard to any factor, including veteran and disability status, and criminal history, in accordance with applicable local, state, and federal laws.

Forterra is also committed to working with and providing reasonable accommodations to individuals with disabilities. Please let your recruiter know if you need an accommodation at any point during the interview process.

To see Forterra's candidate data privacy policy, please visitForterra Legal | Job Applicant Privacy Notice.