Senior Technical Product Manager - Agent Runtime

ulliOwn the agent runtime architecture. /liliProvide technical direction for our agent runtimes: agent orchestration and multi-step loops, context and memory management, retrieval strategy, tool execution and orchestration, model routing and fallbacks, guardrails, caching, streaming, and concurrency. /liliDefine and own the non-functional requirements of the platform including how well the platform runs: latency, throughput, reliability, graceful degradation, scalability under concurrent agent load, token-cost economics, security, and observability/telemetry. /liliOwn the creation and running of product evals that prove our agents are accurate and reliable. /liliOversee the model strategy including model selection, swaps, and provider decisions; cost/performance trade-offs; staying current as frontier models move. /liliTrack market intelligence and provide thought leadership on agentic systems to differentiate the company. /li /ulRequirementsulli4+ years in product management, with 2+ years on technical/platform products (APIs, infrastructure, ML systems, or developer tools) /liliDeep understanding of LLM-based systems and agent architectures: prompting, tool use, planning, multi-step agent loops, context and memory, multi-agent systems - with enough grasp of retrieval/vector search to judge when it's the right tool /liliBackground in distributed systems or infrastructure - you understand how runtimes behave under scale and how a change ripples across the stack /liliTreats non-functional requirements as a first-class product surface: latency, reliability, xgiwjmb scalability, cost, observability /liliHands-on experience creating and running evals that measure and improve agent accuracy and reliability at scale /liliFluency in model operations: model selection and swaps, provider trade-offs, token cost/performance optimization /liliCan write technical specs that engineers review for feasibility (not correctness) and prototype with code to validate hypotheses; comfortable with architecture trade-offs (latency vs. accuracy, cost vs. capability) /li /ulbr