AI Product Manager, Agents & Orchestration

1mind is building an AI-led go-to-market workforce, and this role owns the agent runtime and orchestration layer powering its Superhumans. The AI Product Manager will define how agents plan, use tools, share memory, hand off work, and are evaluated, while turning GTM workflows into a configurable platform.

Responsibilities:

• Own the agent runtime every Superhuman runs on: planning, tool calling, memory reads and writes against the context graph, guardrails, and the Assist, Co-pilot and Autonomous trust ladder. One runtime, many roles
• Design multi-agent orchestration: handoffs between Superhumans across the customer journey (inbound SDR to SE Ride-Along to onboarding to CSM), shared state across a deal or account, and conflict resolution when two agents want to act on the same contact
• Turn GTM orchestrations into a product. Packaged plays by funnel-stage outcome, like qualification and activation or retention and expansion, that customers configure instead of build. You define the primitives: triggers, steps, agent assignments and human checkpoints
• Build and own the eval harness for agent behavior, including task-level success, objection-to-advance rate, escalation precision, and hallucination and tool-error rates. Set the bar that gates every release and own the dashboard that tracks it
• Define the platform surface: the APIs, configuration and admin experiences that app-layer PMs, forward-deployed engineers and enterprise admins use to compose and tune agents, including per-enterprise adapters and fine-tuning hooks
• Run a weekly ship cadence with GenAI and engineering, with enablement and customer success looped in before every release
• Write the specs engineers want to build from (data models, state machines, API shapes and the tradeoffs between them), and prototype in code or no-code tools before you write the doc

Requirements:

• 5+ years of product management, including at least 2 years shipping LLM-based agents to production users. You've owned a product where the model's behavior was the product
• Hands-on with the agent stack: tool calling, planning loops, memory and retrieval, multi-agent handoff patterns, and the ways each one fails. You can read a trace and tell us what went wrong
• You've built or owned an eval harness and used it to gate releases. You think in task success rates, not vibes
• Technical enough to spec data models, state machines and API shapes, and to explain the tradeoffs between them
• Comfortable deciding with incomplete data and shipping weekly. You'd rather fix a v1 in production than polish a v2 in a doc
• Clear writing. Your PRDs are short, your eval reports have numbers, and your Slack messages make sense
• Experience at an AI-native company, or on a large platform team paired with founder or early-stage startup experience
• Fine-tuning or RL exposure: SFT, DPO or reward-based training on outcome-labeled data, or per-tenant adapters
• Voice or real-time conversational agents, where latency and interruption handling shape the product
• GTM domain depth: you've carried a quota, run a CS book, or built for people who did
• Workflow or orchestration products at the platform level
• Enterprise platform experience: admin surfaces, permissions, audit, and the customers who ask for them