Principal Software Architect - Emerging Technologies

Granicus•Remote,

4d•$175,000 - $206,500•Remote

About The Position

Within the Office of the CTO, Granicus operates an AI-native software development lifecycle, a production engineering model where autonomous agents perform high-confidence implementation work and senior engineers orchestrate, review, and own the outcome. This role joins a team focused on scaling this model across additional delivery teams. The engineering standards are aligned with leading software and AI organizations, utilizing practices like generator-verifier architecture, eval-driven development, staged deployment, and high-volume autonomous pull-request pipelines within a FedRAMP-authorized environment. This role is for practitioners who will set and uphold these standards, not just engineers who use AI tooling. Key operating constraints include agents executing only within branches, all agent-generated code passing senior human review before merging to production, and autonomous execution serving as a force multiplier without transferring accountability. Compliance obligations such as NIST 800-53 Rev 5, WCAG, SOC 2, and applicable FedRAMP authorizations are treated as engineering requirements and a source of competitive advantage.

Requirements

Architecture of systems at scale with accountability for outcomes, including distributed systems, CI/CD, and developer platforms where design decisions carried significant operational consequence.
Deep CI/CD and developer-platform expertise, with experience building reliable and performant pipelines for testing, staged deployment, rollback, and observability that other engineers depend on.
Hands-on experience operating LLM agents in production paths, including orchestration, tool use, sandboxing, evaluation, and understanding failure modes at volume.
Security-architecture expertise, including least-privilege design, secrets management, network boundary enforcement, and threat modeling, with work that withstands audit and adversarial review.
A track record of setting standards adopted by others, such as a branching model, review protocol, or platform pattern in active use because you defined it.
The ability to serve as an accountability gate without becoming a bottleneck, designing review tooling and conventions that maintain high signal and low latency.

Nice To Haves

High-assurance security architecture, with demonstrated experience designing and defending a security boundary within FedRAMP, FedRAMP High, defense, financial services, or healthcare, including direct experience with formal audit.
NIST 800-53 Rev 5 fluency, including the ability to map controls to engineering guardrails directly.
AWS Bedrock or GovCloud experience maintaining inference and data within a compliance boundary.
Experience with eval-driven development or generator-verifier architectures at a frontier-model or leading platform organization.

Responsibilities

Own the generator-verifier architecture, enabling volume to scale without quality regression by having generator agents produce changes and verifier agents, automated tests, evaluation suites, SAST/DAST, and policy checks validate them.
Define agent orchestration patterns, including task decomposition, dispatch, sandboxing, and reassembly across the agent array, with secure state and credential handling via WASM-isolated runtimes and encrypted credential vaults.
Build and operate CI/CD for autonomous pull-request volume, a pipeline that triages, tests, and routes a high volume of agent-generated pull requests while keeping human review efficient and high-signal.
Establish eval-driven development as the default discipline by defining measurable acceptance criteria so changes are validated against evaluation suites rather than subjective judgment.
Own the security boundary, ensuring in-VPC inference via AWS Bedrock (including GovCloud), least-privilege agent permissions, and an explicit mapping from NIST 800-53 Rev 5 controls to concrete pipeline guardrails, enforcing compliance through architecture.
Serve as the final human accountability gate, defining and enforcing the requirement that no agent-generated code reaches production without senior human review, supported by review tooling that maximizes signal.
Define the engineering standards inherited by every team, including the branching model, review protocol, evaluation conventions, and staged deployment, and maintain the reference implementation the organization builds on.
Govern graduated autonomy by defining the criteria by which a workstream advances from supervised to autonomous execution, supported by measured agent reliability rather than assumption.