SlipstreamJobs tracks this role from the company's public career site. Apply directly on the employer's site.
Supabase is hiring a Staff AI Platform Engineer to build the execution layer for the company's internal AI operating system. This is a foundational role where you will be the sole engineer responsible for designing and shipping a production-grade agent platform from the ground up.
You will own the complete architecture and implementation of an event-triggered queue, a headless model-agnostic runtime, durable state management, human review gates, atomic rollback capabilities, and comprehensive logging of all prompts, tool calls, and decisions. Beyond the runtime, you will build the evaluation layer that makes agents trustworthy through golden test suites with behavioral assertions, rubric-based judging, safety cases, and CI gates that prevent regressions.
A core responsibility is enforcing governance in code rather than policy. You will design risk-tiered agent capabilities where dangerous operations have no code path to execute them, implement least-privilege credentials per agent, create tool-permission gates, and maintain a decision audit log. Critically, agents will never be able to autonomously write commitments (owners, due dates, statuses) into shared work systems—this constraint will be structurally enforced through your platform design.
You will build and register the agent portfolio across executive reporting, team-lead operations, individual-contributor support, and a meta-layer that observes the platform itself and files improvements. You'll design how the system contacts people, balancing interruption budgets and message bundling to drive adoption. You will own all platform tooling including compilers, validators, and distribution paths that embed agents into repositories and chat surfaces.
The role requires recursive thinking: building generators and inventory systems rather than individual artifacts, and applying the same philosophy to evaluation—building the machine that decides whether every future agent is allowed to ship. You will also think inversively, starting from failure modes and working backward to designs that make violations structurally impossible. Finally, you will instrument the platform's return, computing operating measures and a monthly value ledger so the system's impact is measured rather than claimed.
