Staff Platform Manager, Conversational Products

Airbnb

3d•$200,000 - $240,000•Remote

About The Position

Airbnb is a vision and mission driven company, and our Product Managers embody that mindset. Platform PMs imagine the ideal end state for our community first, work backwards from it, and deliver it in a scalable way. The AI Assistant team owns the agentic AI system that powers Airbnb's support experience for millions of guests and hosts. This is one of the highest-visibility applied-AI efforts at the company, and it sits at the intersection of large language models, platform thinking, and a genuinely two-sided marketplace. You will own the platform that determines how our AI Assistant reasons, retrieves, and responds: the layer that interprets each request and routes it, the knowledge and capabilities the assistant draws on, the actions it can take to actually resolve an issue, and the evaluation systems that keep it safe and accurate at scale. You will own what a good outcome looks like for the user, and the criteria we measure it against, working through partners who own the underlying knowledge and the engineering implementation. This is a role where Product both directs technical work and does it themselves. You will diagnose architectural problems, author the artifacts that become production behavior, and help drive engineering and data science decisions alongside your key partners. Support is high-stakes: people reach out when something has gone wrong, often with another party involved. You will be responsible for making those moments accurate, safe, and genuinely helpful, increasing how often the assistant fully and correctly resolves a request on its own while expanding coverage across more problem types and user touch points.

Requirements

9+ years building technology products, with at least 5 in product management or a closely related technical role (engineering, data science, or applied research) from which you owned product direction.
Direct experience shipping generative AI or ML products to production, ideally including retrieval-augmented generation, agentic or function-calling architectures, and evaluation systems.
Hands-on technical depth. You read and debug prompts, understand ML and engineering constraints, and are comfortable personally authoring the artifacts that shape production behavior.
Data fluency. You can work directly with production data, validate an analysis end to end, and know when a result looks wrong, whether or not you write the query by hand.
Depth in LLM evaluation. You have designed, calibrated, or certified automated model-based evaluations, and can turn a vague quality signal into diagnostic, measurable components.
A strong sense for great user experience, including how latency, tone, and trust shape an AI interaction.
Comfort owning an ambiguous, cross-cutting mandate that spans several teams' focus areas, and the influence to drive alignment accordingly.
A track record of scaling your impact through others: setting a clear bar, assigning owners, and holding them to it.
Strong executive communication: can lead with the decision and the ask, and adjust the narrative to the audience.
Radical thinking paired with strong execution: can envision a system meaningfully better than today and articulate the path to get there.

Responsibilities

Set the product direction for how the assistant reasons and responds, and paint a multi-quarter vision with the customer at the center; align that vision with senior leaders and cross-functional partners.
Work fluently with production data. Explore the data directly, use modern AI tools and coding agents to move fast, validate an analysis, and dig in yourself to tell when a result doesn't look right.
Build new capabilities end to end, from spotting the need through the hands-on work that makes the behavior real and gets it calibrated with human input.
Decide what a correct, complete resolution looks like for each kind of user problem, and get engineering, policy, and knowledge partners aligned around that definition.
Define what success means in measurable terms: how often issues get fully resolved, how accurate and safe the answers are, and the bar that a change needs to clear in order to launch.
Own launch readiness for major model and platform changes, balancing speed to learn against the safety and risk work that has to happen before anything scales.
Diagnose why a complex AI system is failing, and judge where the fix actually belongs.
Own the evaluation strategy: LLM-based evaluators (LLM-as-judge), offline and live-traffic evaluation, calibration and certification, and the tooling that lets non-engineers safely improve the assistant.
Bring teams with different perspectives to a shared answer, set agreements with partners early, and push for a single source of truth across the product.
Present to leadership regularly, leading with the decision, the tradeoffs, and the ask, and tailoring the narrative to what the audience cares about.