Principal Conversation & Agentic Experience Architect

Twin Health · Remote, USA · Engineering

Posted 2026-11-20

Apply for this role →

Opportunity

Twin Health has built a powerful system of specialized intelligence around each member: a Digital Twin, specialist AI agents, clinical care teams, and human coaches. As we scale, our challenge is no longer building capable individual agents — it is making sure members experience one Twin: a single, coherent, trustworthy health companion, no matter which agent, coach, or clinician is behind a given moment.

We're hiring a Staff-level architect to own how Twin's AI agents understand members, speak, collaborate, personalize, and escalate — and to encode those standards not just as documentation, but as executable systems (Skills, agent templates, prompts, evaluation suites) that our product, clinical, and AI teams build on every day.

This is the "model behavior" and "conversation architecture" function that frontier AI labs have established for their assistants — applied to the highest-stakes, most personal domain there is: human health and metabolic disease reversal.

Responsibilities

Define the North Star for Twin intelligence

Codify the principles governing every intelligent interaction across the Twin ecosystem: what Twin believes about health, motivation, personalization, autonomy, encouragement, intervention, and human care.

Translate philosophy into behavioral specifications precise enough to build, evaluate, and enforce.

Own the voice and behavior system

Design and maintain Twin's voice, tone, and personality system — empathy, encouragement, explanation, insights, goal setting, intervention, and escalation — as a living, versioned standard.

Ensure agent-to-agent consistency so members never feel organizational or technical seams.

Architect multi-agent orchestration and handoffs

Map the specialist agent ecosystem: roles, responsibilities, boundaries, context requirements, and overlaps.

Define when an agent acts, listens, hands off to another specialist, or brings a human (coach or clinician) into the loop — and design the context-preservation protocols that make those transitions seamless.

Build personalization and trust frameworks

Define how interactions adapt to member goals, state, history, progress, and preferences.

Establish standards for safety, privacy, consent, clinical boundaries, transparency, and healthy AI interaction — including guarding against dependence, confusion, and over-engagement.

Make the architecture executable

Partner with AI engineering to encode standards as reusable Claude Skills, specialist agent templates, system prompts, and governed context/MCP integrations.

Build and own the evaluation and regression suite: representative scenarios and criteria that continuously test whether agents remain aligned with Twin's North Star as models, prompts, and experiences evolve.

Establish governance: how new agent behaviors and Skills are created, reviewed, versioned, and approved.

Lead across the organization

Operate as the connective tissue across product, design, content, clinical, engineering, and leadership — without direct authority, through clarity of vision and quality of artifacts.

Run workshops and enablement so teams can create new specialist experiences without reinventing Twin's foundational intelligence each time.

Other duties as assigned.

What Success Looks Like (First 12 Months)

A ratified Twin Agent North Star and Voice + Behavior System in active use by every team shipping agent experiences.

A working agent handoff architecture deployed across at least the top member journeys, with measurable reductions in experience seams (drop-offs, repeated context, conflicting recommendations).

A Skill library and specialist agent template adopted as the default starting point for new agents.

An evaluation framework running against every model, prompt, or Skill change before it reaches members.

Clinical, privacy, and safety stakeholders who trust the system — and can point to where their requirements live in it.

Qualifications

8+ years in conversation design, content design/architecture, UX for AI, model behavior, or adjacent fields — with 3+ years shaping LLM-based product experiences in production.

Demonstrated Staff-level scope: you've defined behavior/voice/interaction systems adopted across multiple teams or product surfaces, not just individual flows.

Exceptional writing and taste. You can craft a single model response that lands perfectly — and a specification that lets a hundred responses land the same way.

Hands-on fluency with modern LLM tooling: prompt and system-prompt design, agent frameworks, tool use / MCP-style integrations, evaluation design (automated evals, human evals, rubric design). You don't need to be an ML engineer, but you must have real technical intuition for how prompts, context, and data change model behavior.

Systems thinking: experience designing multi-agent, multi-surface, or multi-channel experiences where context, state, and persona must persist across boundaries.

Comfort with ambiguity and philosophy-to-practice translation: you enjoy hard questions about autonomy, trust, and healthy human-AI interaction — and always drive them to concrete, testable standards.

Strong cross-functional leadership: proven ability to align research, engineering, product, design, and domain experts (in our case, clinicians) around a shared behavioral standard.

This remote opportunity based out of the U.S. Applicants must be authorized to work for any employer in the U.S. We are unable to sponsor or take over sponsorship of an employment Visa at this time.

Preferred Qualifications

Experience in healthcare, digital health, coaching, or another regulated / high-trust domain (HIPAA familiarity a plus).

Experience with safety-critical model behavior: escalation design, crisis pathways, clinical boundaries, refusal/deferral patterns.

Familiarity with Anthropic's Claude ecosystem (Skills, MCP, agent SDKs) or equivalent (OpenAI Assistants/Agents, Gemini).

Background in linguistics, HCI, behavioral science, service design, or content strategy.

Experience building or operating LLM evaluation infrastructure and data flywheels from production signals.

Compensation and Benefits

The compensation range for this position is $220,000 - $240,000  annually.

Twin has an ambitious vision to empower people to live healthier and happier lives, and to achieve this purpose, we need the very best people to enhance our cutting-edge technology and medical science, deliver the best possible care, and turn our passion into value for our members, partners and investors. We are committed to delivering an outstanding culture and experience for every Twin employee through a company based on the values of passion, talent, and trust. We offer comprehensive benefits and perks in line with these principles, as well as a high level of flexibility for every Twin

A competitive compensation package in line with leading technology companies

A remote and accomplished global team

Opportunity for equity participation

Unlimited vacation with manager approval

16 weeks of 100% paid parental leave for delivering parents; 8 weeks of 100% paid parental leave for non-delivering parents

100% Employer sponsored healthcare, dental, and vision for you, and 80% coverage for your family; Health Savings Account and Flexible Spending Account options

401k retirement savings plan

Salary range for US jobs

US Salary Range

$220,000—$240,000 USD

Apply for this role →

← Back to all jobs