Staff AI Infrastructure Engineer

Reign is hiring a Staff AI Infrastructure Engineer to build and operate the platform that turns conversations and specifications into verified pull requests. The role owns agent orchestration, verification pipelines, managed services, compute and sandbox environments, and safe infrastructure guardrails.

Responsibilities:

• The factory: conversation to production. Any conversation can become a verified PR in a sandbox before the conversation ends. You take the overnight loops, the pipeline, and the handoff tooling our engineers built for themselves and make them the system: an orchestrator holding judgment and the team's context, cheaper models executing with exact instructions, success criteria at every step, the repo updated as it goes. Built on your own machine until it runs overnight against success criteria, then put in the cloud and replicated
• Verification that can see under the hood. Screenshots and video already cover UI. Checkout, billing, and payment internals cannot be verified by looking at a screen. You build the classic answer with modern parts: AI-written tests living in a real CI pipeline, sandboxes seeded with synthetic or sanitized data so a full sweep never touches production, parity checks against the analytics store, and experiment arms as release gates. The agent harness that runs outside CI today runs inside it
• Design first, then the last mile. Every build starts with the agent proposing how it will build, split, and ship the change, and an engineer approves that design as part of the spec gate. Sit with the co-founders and product owners who build with agents. Find exactly where their work stalls between working and merged. Build the step that carries it through, whether that is upfront planning, a review lane, a verification gate, a data fixture, or a skill the agent loads. Then measure how many of them ship each month and make the number go up
• Memory and the watcher, as services. The team brain (Postgres with vector and full-text search) and the watcher (a service that reads every log, metric, and pipeline, reports causes rather than symptoms, and drafts the fix) need to run 24/7 as managed services with an uptime number. You own the runtime, data, and access for both. The engineers who built the logic keep the logic
• Compute and environments. The runner fleet for agent swarms, cloud development environments with safe data, per-sandbox databases, and an isolated team AWS sandbox with real controls. Cloud, bare-metal, or on-prem decided with usage and cost data, and revisited as the numbers change
• Same system, same power. Every engineer, onshore and offshore, runs the same orchestrator on the same tier of account. You own the accounts and usage telemetry, the non-production vault, sandbox plus synthetic data, time-boxed production access, and a model supply that survives spend caps and provider changes, including a router for isolated workloads on other providers
• Guardrails that keep it safe. Production and shared infrastructure change only through Terraform PRs; inside the isolated sandbox account, agents provision what they need under policy-as-code guardrails, with no tickets. Automation proves itself in a sandbox before a PR exists. Production only ever sees a reviewed merge. Security work never runs unattended. No quota layer on top: we build the loop, not the whip