You will own the scientific agenda. You will convert a fuzzy behavioral goal into an end-to-end plan that covers data, agent architecture, evaluation, and the product experiment that tests it. You will write code every week and build large parts of the experimental stack yourself, because it does not exist yet.
The first focus areas are agentic products where a domain expert holds judgment the system cannot replace. Amazon Connect places an assistant beside a person handling a live conversation. AWS Bio Discovery places one beside a scientist running experiments. Both need evaluation that measures collaboration quality rather than task completion alone.
The role changes shape as the Lab matures. It starts as hands-on science in close partnership with product and engineering. Later it moves inside a product team to carry adoption of what the Lab proved.
Key job responsibilities
- Own the scientific strategy for human-AI collaboration across agentic and multimodal systems.
- Convert desired interaction behaviors into falsifiable hypotheses, evaluation tasks, and system requirements.
- Build the evaluation system that measures teammate behavior, trust, adaptation, human contribution, and failure recovery.
- Design agent systems that use memory, tools, planning, and recovery, then test them with the people who do the work.
- Design data collection and curation for language, speech, and interaction traces.
- Build significant parts of the experimental stack yourself, because that stack does not exist yet.
- Diagnose failures across data, models, orchestration, evaluation, and product interaction.
- Define the requirements engineering needs to turn a proven method into a product capability.
- Partner with design, product, engineering, and behavioral research from problem definition through product validation.
- Run experiments with partner product teams, then report what worked, what failed, and what changed as a result.
- Set the standard for reproducible experiments, evidence, and scientific review inside the Lab.
- Mentor scientists and engineers without moving away from hands-on work.
- Represent the work in internal reviews and in appropriate external scientific venues.
A day in the life
Your week has two centers of gravity.
Most days you build. You take a claim about how a teammate should behave, design the smallest experiment that can falsify it, and run it. You read interaction traces from real sessions, then argue with the engineers about what they mean.
The rest of the week belongs to the people the work is for. You sit with a contact center agent or a bench scientist and watch where the system helps and where it intrudes. You leave with the next hypothesis. Often you leave with evidence that kills the last one.
About the team
The Lab is a lead who still writes code, an Applied Scientist, and engineers. Applied AI Solutions leadership approved it with one instruction, to explore the limits of what we can build, and one number, to ship six or more months ahead of anyone else.
You own the agenda, the hypotheses, and the evaluation. The engineers build the systems that test them, then the products that carry the ones that survive. Neither seat hands work over a wall.
We have a sunset clause. Once a capability is proven and integrated into its product, the Lab moves to the next frontier.
BASIC QUALIFICATIONS
- PhD in computer science, machine learning, artificial intelligence, or a related technical field, or a Master's degree with equivalent applied science experience, or an equivalent body of work
- Experience developing agentic AI systems or large language model applications through an end-to-end product cycle
- Experience designing evaluation for systems that have no standard benchmark
- Experience writing substantial research or production code in Python or a comparable language
- Experience leading complex scientific work across engineering and product partners
- Evidence of independent decisions in ambiguous, consequential technical environments
PREFERRED QUALIFICATIONS
- Experience with post-training methods, including supervised fine-tuning and reinforcement learning
- Experience with multimodal models across language, speech, and vision
- Experience with speech systems or real-time conversational systems
- Experience measuring collaboration quality, trust, adaptation, or human contribution in AI systems
- Experience building agent systems that use memory, tools, planning, and recovery mechanisms
- Experience founding a scientific program, or joining an early team before that team had established its methods
- A record of scientific influence through publications, patents, open-source work, or deployed systems
- Experience mentoring senior scientists and engineers without moving away from hands-on work
- Experience partnering with design, product, and behavioral research from problem definition through product validation
- Experience taking a proven method into a product team and staying until that team adopted it
The base salary range for this position is listed below. Your Amazon package will include sign-on payments and restricted stock units (RSUs). Final compensation will be determined based on factors including experience, qualifications, and location. Amazon also offers comprehensive benefits including health insurance (medical, dental, vision, prescription, Basic Life & AD&D insurance and option for Supplemental life plans, EAP, Mental Health Support, Medical Advice Line, Flexible Spending Accounts, Adoption and Surrogacy Reimbursement coverage), 401(k) matching, paid time off, and parental leave. Learn more about our benefits at https://amazon.jobs/en/benefits.
USA, TX, Austin - 198,900.00 - 269,000.00 USD annually
