Responsibilities

• Discover and define workflows: Work directly with sellers, product marketing managers (PMMs), and campaign managers to map how they work today, where time is wasted, and what good output looks like. Begin with direct observation and conversations with the people doing the work, then turn those insights into clear technical requirements.
• Define the agent runtime and application architecture: Design how the system manages context, state, tools, permissions, retrieval, evaluation, and human review. Define how it ingests account signals and playbook intelligence, routes requests to specialist agents, and surfaces outputs in the tools teams already use.
• Design trusted knowledge flows: Define how the system finds, evaluates, and uses account, persona, product, competitive, and campaign information. Make thoughtful decisions about source-of-truth systems, information freshness, conflicting sources, structured and unstructured data, and provenance.
• Build routing and specialist agents: Build the routing and execution system that selects the right tools, data sources, and specialist agents for each request. Cover sales asset generation, campaign briefs, competitive responses, copy production, and field support for active deals.
• Integrate with Sales AI and go-to-market platforms: Connect the AI layer to the platforms 1Password runs on, including Salesforce, Seismic, Gong, Notion, Snowflake, and Slack. Ensure it works alongside existing Sales AI tools and workflows, with permission-aware tool use and clear boundaries for what each agent can access and change.
• Design for trusted access and safe autonomy: Design and test permission-aware retrieval, source-level authorization, audit trails, provenance capture, prompt injection defenses, retention and deletion controls, and safe handling of customer and account information. Define where agents can act independently, where human approval is required, and how the system records those decisions.
• Evaluate and improve agent behavior: Tune agent behavior using representative evaluation sets, seller feedback, usage data, and output quality signals. Measure accuracy, relevance, consistency, editing effort, adoption, and other workflow-specific outcomes. Improve prompts, tools, retrieval, models, and agent behavior as the go-to-market motion and product landscape evolves.
• Build quality and reliability guardrails: Build evaluation frameworks, regression pipelines, release checks, and escalation paths that keep outputs accurate, useful, on-brand, and consistent. Define how workflows handle uncertainty, failure, and human review. Production output is valuable only when it is trustworthy and useful to the teams receiving it.
• Operate and evolve the system: Own the roadmap and the full lifecycle of the systems you build, from prototype through reliable production use and eventual retirement. Monitor system performance, usage, quality, latency, and cost. Identify what to build next, deprecate what is not working, and partner with the appropriate teams to resolve failures and improve reliability.

Requirements

• Production AI systems: 6+ years experience designing, building, and operating workflows that use large language model (LLM) APIs, tool calling, retrieval, structured outputs, and agent workflows. Candidates should be able to describe systems they have shipped, how they evaluated them, what failed, and how they improved them based on real user feedback.
• Agent-system architecture: Experience designing workflows with clear agent boundaries, context and state management, permission-aware tool use, evaluation, observability, failure handling, and human escalation. Familiarity with tools such as Claude, Codex, Cursor, Dust, LangChain, or LangGraph is useful but not required.
• Technical judgment: Choose the simplest reliable approach for the problem. Know when to use deterministic code, a workflow, retrieval, a single agent, multiple agents, or human review. Avoid adding agentic complexity where it does not create meaningful value.
• Builder's instinct: you'd rather put a working prototype in front of a seller today than write a spec about what's theoretically possible next quarter.
• Evaluation: Experience building evaluation frameworks to measure agent output quality through prompt evaluations, output scoring, and regression testing against verified examples.
• User-centered delivery: Prototype quickly with users, learn from real feedback, and turn what works into documented, reliable production systems.
• Cross-functional collaboration: Translate workflow needs and feedback from PMMs, sellers, and campaign managers into clear technical requirements.
• Technical communication: Explain system architecture and technical tradeoffs to engineers, then translate workflow constraints and user feedback into clear decisions with PMMs and sellers.

Preferred

• Go-to-market fluency: Working knowledge of business-to-business software as a service (B2B SaaS) marketing and sales workflows, including personas, buying stages, positioning, competitive narratives, campaigns, sales enablement, and content governance. Able to translate feedback from PMMs, sellers, and campaign managers into clear product requirements and measurable improvements.
• Sales, marketing, and data platforms: Familiarity with Seismic, Salesforce, Gong, Notion, Snowflake, or Slack APIs; experience with integration platform as a service (iPaaS) tools for cross-system automation; exposure to B2B SaaS or enterprise go-to-market environments.

Benefits

• Health and wellbeing
• 👶 Maternity and parental leave top-up programs
• 🩺 Competitive health benefits
• 🏝 Generous PTO policy
• Growth and future
• 📈 RSU program for most employees
• 💸 Retirement matching program
• 🔑 Free 1Password account
• Community
• 🤝 Paid volunteer days