About the role
Responsibilities:
- Integration generation at scale. We don't hand-write connectors. Build the pipeline that generates them from whatever a vendor publishes, which is often not a clean OpenAPI spec—and own spec normalization, pinned provenance, and deterministic regeneration across a growing catalog.
- Tool surfaces an AI can actually call. An action's name, description, schema, and examples are the product surface a model sees. Design them so the assistant picks the right action first try, scope them so they are safe for an AI to call, and measure whether it does.
- Permissions and credentials you can prove. If a file is private to HR, a marketer must not see its title, its URL, or that it exists. Own access-model mirroring and delete detection, and pick the smallest credential policy each provider needs across validation, reconnect, revocation, and tenant isolation.
- AI-native development itself. Agentic coding is how this team ships, not a productivity add-on. Author the skills and guardrails that let agents do integration work correctly, keep the deterministic provider-fake harness that makes fast iteration possible, and review generated code critically before anything ships.
- Engineering leadership. Raise the technical bar through architecture decisions, code review, and coaching. Partner with product and design to shape requirements and explain tradeoffs clearly.
Requirements:
- Experience leveraging or critically thinking about how to integrate AI into work processes, decision-making, or problem-solving. This includes current, hands-on fluency with AI-assisted and agentic development.
- 8+ years of software engineering with strong fundamentals in system design, API design, relational data modeling, and testing.
- Hands-on depth building and operating third-party integrations that real tenants depended on—webhooks, rate limits, pagination, incremental sync, and delete detection.
- Experience designing API or SDK surfaces that another system consumes correctly without a human reading the documentation—and interest in doing that for a model.
- Ownership of a versioned contract where a rename breaks a customer: stable identifiers, deprecation, and migration.
- Sound judgment about least-privilege access—scope minimization per action, admin consent, secrets handling, tenant isolation—and about which actions an AI may take on its own.
- Comfort with ambiguity and the drive to build from 0-1 without a playbook.
- A record of technical leadership: architecture ownership, code-quality bar-raising, and mentoring engineers.
Nice to have
- Experience evaluating tool or function selection—measuring whether a model picks the right action.
- Experience with RAG or the Model Context Protocol (MCP).
- Familiarity with SCIM or directory sync.
- Background in IT service management or employee support products.