AI Platform Engineer

The role

This role sits on a newly formed central team that is designing the company's AI-driven internal operations, where AI handles a large share of routine operational work. The engineer is the sole member of the platform group and is expected to settle open architecture questions and own the full stack. The core concern is agents that operate under tiered risk levels, with review and approval scaled to risk and certain actions blocked at the code level.

What you would do

• Build and launch the production agent runtime: event-driven intake, a model-neutral execution engine, state that persists through crashes, human sign-off steps, reversible writes, and complete run logs.
• Create the measurement framework that judges agent quality, with fixed test sets, behavioral checks, scored rubrics, mandatory safety tests on every run, and a merge gate that stops regressions.
• Develop and register the agent lineup covering reporting, drafting, linting, triage, Q&A, and a monitoring agent that finds and fixes platform issues.
• Hard-code access restrictions so each agent holds scoped credentials and actions outside its tier have no execution path, with audit logs of consequential decisions.
• Design outreach to staff using per-person interruption caps, grouped notifications, and delivering something useful before asking for input.
• Maintain platform tooling: the build-and-validate pipeline, registry data consistency, and the mechanisms that push context and capabilities into code repos and chat tools.
• Build pipelines that convert raw system events into per-team maturity scores, with query-level traceability so results can be checked and contested.
• Set up a ledger recording the work each agent takes on, and compute a monthly value figure from measured data.
• Take ownership of infrastructure from provisioning through deployment, monitoring, and rollback, and wire the platform to report on its own performance.

What they asked for

• Track record of running LLM-based agent systems in production that others relied on, with real usage and at least one incident you can walk through.
• Experience designing evaluations rather than ad hoc checks: golden datasets, behavioral assertions, judging rubrics, pass thresholds, and CI gating.
• Hands-on depth with the APIs of systems where work happens, plus experience authoring MCP servers and an understanding of their failure modes.
• Competent building and running systems in Python on GCP, including a cloud data warehouse and declarative provisioning, with the ability to deploy, monitor, and revert your own systems.
• Comfortable making architecture calls independently rather than routing them elsewhere.
• Thoughtful about how software interrupts people, treating each notification as a limited resource to be spent carefully.

Nice to have

• Public contributions in the field, such as an open-source agent framework, MCP server, evaluation harness, or widely referenced writing on agent dependability.
• Familiarity with tracing agent runs, attributing cost per run, and turning raw logs into reports people act on.
• Prior experience building an internal tool that non-technical colleagues chose to use, and an ability to explain changes made after observing how they used it.