Staff Product Manager, Agentic AI Applications

Location: San Francisco or Mountain View, California GAQ327R225 Databricks is building an Agentic Enterprise applications Platform a scalable, governed AI application platform built on Databricks that enables any internal team (GTM, Finance, HR, Legal, Product) to build production grade agentic applications in weeks, not months. The platform provides a managed agent runtime, standardized MCP connectors to enterprise systems of record, a shared UI component library with a design system, an intelligence data/context layer and a gold-standard promotion pipeline from prototype to production. As a Staff Product Manager , you will own the product strategy, roadmap, and execution for the Agentic Platform the foundational layer that every domain workspace depends on.

You will work across agent runtime, MCP connectors, intelligence layer, evaluation framework, developer experience, and governance to deliver a platform that reduces time-to-production from months to weeks while maintaining enterprise grade quality, security, and reliability. You will partner closely with Application Engineering, the CIO organization, and domain teams across GTM, HR, Finance, and Product to ensure the platform serves real needs and scales with the organization. The impact you will have: Own the Agentic Platform strategy and roadmap.

Define what ships, in what order, and why. Translate organizational outcomes into concrete platform capabilities with measurable success criteria. Define and drive the agent and runtime.

Establish the managed agent runtime supporting multi step orchestration with durable execution, model gateway abstraction across all providers, governed tool invocation, and configurable per-agent guardrails (cost ceilings, timeouts, blast radius limits). Build the MCP connector ecosystem. Own the strategy for standardized, bidirectional connectors various systems of record.

Drive on behalf of identity propagation, idempotency, dry-run/preview mode, and a connector SDK that lets domain teams onboard new systems without platform changes. Establish the intelligence layer. Define the three layer data architecture: knowledge graph (curated domain knowledge), context graph (live entity state from systems of record), and temporal memory (session, user preferences, and episodic history).

Ensure unified retrieval across vector, structured, and graph sources with source traceability on every context element. Ship the evaluation and quality framework. Own the AI-judge evaluation pipeline: offline eval with golden datasets, online LLM-as-judge scoring, domain-specific judges (Finance, HR, Legal, Sales), and mandatory evaluation gates in CI/CD.

No agent reaches production without passing quality and safety thresholds. Design the developer experience. Make the platform self-service by construction.

Domain teams provision agent projects, promote across environments, and access connectors without platform-team tickets. SDK, CLI, sandbox environments, agent templates, and documentation — the paved road must be faster than building bespoke. Target: idea to production in 4 weeks for a standard agent.

Define the federation and adoption model. Establish the three tiers of adoption (platform built, domain built on platform, citizen developer edge apps) with governance checkpoints at each gate. Drive the gold-standard promotion pipeline from edge prototype to production hardened service.

What we look for: 8+ years of product management experience, with at least 3 years on internal platform, infrastructure, or developer-experience products. Deep experience building platforms that other teams build on you understand the difference between a platform and an application, and you have opinions about API design, developer ergonomics, and self service. Demonstrated experience with AI/ML platforms, agent frameworks, LLM-powered applications, or agentic systems.

You know what an agent runtime is, what RAG means in practice, and why evaluation is the hardest part.