Summary
What you’ll impact
The Founding Engineer will be the first engineering hire on the delivery team, building and operating AI-native production agents and workflows for hourly workforce customers. This forward‑deployed role works directly with design customers to map real workflows, create evaluation harnesses, human‑review layers, and turn bespoke solutions into reusable products while handling incidents end‑to‑end.
Responsibilities
What you'll do
- You will own the components: whether they work, how they fail, and how fast bespoke work becomes reusable.
- Build and operate production agents and agentic workflows: orchestration, tool use, data connections, deterministic logic, state, and permissions.
- Work forward-deployed with design customers—sit in their workflow, map the messy reality, and build against live data.
- Build the evaluation harness: eval datasets from real customer work, pre-launch thresholds, regression suites, and post-launch monitoring.
- Build the human-review layer: exception queues, review tooling, escalation paths, and the tracing that makes every agent decision auditable.
- Run shadow-mode parallel runs against the customer's existing process or provider and drive discrepancies to zero before go-live.
- Turn customer-specific work into reusable agents, integrations, and playbooks—productization is the default, customization the exception.
- Handle production incidents end to end, and feed every failure back into evals and controls so it cannot recur.
- Help set the technical bar as the team grows.
Requirements
What you’ll bring
- You have personally built and operated at least one AI agent or agentic workflow in production. You can clearly explain how it worked, where it failed, how you evaluated it, and what you kept deterministic or human-led.
- Strong engineering judgment across models, orchestration, APIs, data and integrations, permissions, observability, and evaluations.
- You have worked directly with customers or end users—discovery, implementation, or incident resolution—and communicate clearly with non-technical people.
- You can break a messy workflow into deterministic steps, agentic decisions, human judgment, and exception paths—and define how each part will be tested.
- You care about keeping plausible but wrong AI output away from customers: grounded context, constrained actions, verification, evals, review, monitoring.
- You thrive in zero-to-one ambiguity: shipping weekly, owning outcomes, and doing whatever the service needs—including unglamorous review tooling and data plumbing.
- Exposure to a regulated domain (payroll, payments, tax, benefits, labor compliance) is a plus, not a requirement.
- Demonstrated production experience matters more than title or years.