The opportunity
As a Staff Product Manager on Datadog's Self-Improving Products team, you will take a new product family from zero to one: a closed loop that watches how a customer's software and its users behave, ranks what is worth changing, proposes the change, proves whether it worked, and…
What you'll do
Set product direction for a new product family by defining the problems and: desired outcomes, and partnering with engineering to find the narrowest first slice that proves the loop closes and delivers value to customers in phases.
Own the product decisions at every stage of the loop: what signal is trustworthy enough to trigger an automated change, which opportunities rank highest and with what confidence, what a proposed change must contain before a customer merges it, what evidence proves it worked, and what the system remembers for the next cycle.
Define the autonomy ladder and its limits, from suggestion, to draft, to: auto-opened pull request, to auto-rollout, along with the controls, defaults, and audit trail customers need to rely on it. Some rungs stay off by design, and you own those decisions.
Set the quality bar for a product that is non-deterministic: build the evaluation sets with engineering, decide what "good enough to ship" means, and hold that bar when it moves a date.
Recruit the first design partners yourself, sit in their triage rotations,: turn what you learn there into the roadmap, and keep talking to them after launch.
Develop and defend the sizing, impact, and cost-to-serve analysis that: engineering leadership and pricing partners need to make resourcing calls, partnering with engineering on the unit economics of inference and the data platform and bringing that cost profile into packaging, metering, and pricing before launch rather than after.
What they're looking for
- Deliver concise written and verbal communication to executive, engineering,: and customer audiences, clarifying complex architecture, quality trade-offs, and go-to-market, including where agentic observability, product analytics, and experimentation are heading and how that shapes the product.
- You have shipped AI products, not prototypes: you have taken at least one AI-native product to production customers, and you can describe what it got wrong in the field and what you changed in response.
- You treat evaluation as product work: you have built or commissioned evaluation sets, set the bar for shipping a model or prompt change, and held that bar under schedule pressure.
- You are fluent in agent failure modes and design for non-determinism: deliberately, accounting for confidence, reversibility, blast radius, review surfaces, and graceful failure.
