The role
From Amazon's own posting.
Amazon is building its own benefits platform, and it is going global — and we are building it on an AI Native stack designed to achieve recursive self-improvement (RSI): a system that doesn’t just ship features but continuously evolves itself, proposing its own next versions from production signals while humans own the judgment gates. This has never been done at this scale on a people-data platform, and the engineers who join now will be the ones who make it real.
For most of Amazon’s history, the systems that decide what employees are eligible for and how they enroll were run by third-party vendors — and outside the US, they still are. Over the next two years we are changing that: supporting more than 2MM employees and their family members WW on Amazon’s own platform, extending retirement and savings to nearly half a million employees who have never had a real system for it, and bringing entirely new populations onto Amazon benefits for the first time.
The benefits you build will be the ones your coworkers and their families actually use — this isn’t abstract infrastructure.
BXT is expanding to build it, and we are hiring Software Development Engineers to build the platform underneath all of it.
You would be joining one of PXT’s largest platform programs while the core is still being designed. The platform we run today was built for one country, and every new plan, carrier, or market has meant an engineering project. What we are building instead is a configuration-driven, AI-native core: a global benefits engine where a country’s rules, a new plan, or a new vendor are configured rather than coded, and where the people who understand benefits can author those rules directly while AI drafts them from the plan documents and contracts. That is not a refactor of something finished — most of it does not exist yet. We are in the first 18 months of a multi-year build. You will make the foundational decisions, not inherit them.
What You Will Build
A global benefits platform. The core enrollment, eligibility, and employee-profile capabilities that let Amazon deliver benefits in any country — where “add a market” means configuring its rules, not spinning up a bespoke build. First countries go live during the initial waves; the architecture you help design determines how quickly the rest follow.
The AI-authoring layer. Systems that turn plan documents, carrier contracts, and a country’s regulations into working benefits configuration — with AI drafting the rules, generating the tests that validate them, and predicting the full population impact of a change before it is ever enacted. Engineers stop being the bottleneck between a new benefit and the employees who get it. Today, onboarding a new country’s benefits rules takes weeks of engineering. With the AI-authoring layer you’ll build, a benefits analyst uploads a plan document and the system drafts the configuration, generates validation tests, and predicts population impact — reducing weeks to hours.
Automation that acts at population scale. Platform capabilities that carry out benefits actions on behalf of hundreds of thousands of people at once — enrollment, coverage changes, annual rollover — with the impact prediction, phased rollout, and safety controls that make operating at that scale confident rather than nervous.
The recursively self-improving loop. This is what makes this role unlike anything else at Amazon today. You will contribute to building the always-on agents and feedback systems that watch how the platform runs in production — what fails, what’s slow, where customers struggle — and autonomously propose the next improvement. Those proposals flow through a common intake where humans approve or decline, and then the build loop ships the change. The system improves itself, run after run, with humans at the judgment gates. No team at Amazon has closed this loop on a people-data platform at this scale. You will be among the first to do it — with the support of senior engineers, rigorous design reviews, and layered safety mechanisms that make bold moves responsible ones.
New benefits domains. Retirement and savings moving onto the shared platform for the first time — including international programs across 30+ countries that run on spreadsheets and country-by-country vendors today. Coverage that starts on day one for new hires instead of weeks later. Whole categories of employee benefit that have never had a real platform behind them.
The engineering foundations that make the pace possible. Continuous delivery end to end, simulation testing that validates a change against realistic populations at machine speed (not a handful of test identifiers), automated test generation, and business rules expressed in a machine-readable form so correctness can be checked automatically instead of living in people’s heads. Built on DynamoDB, EventBridge, Step Functions, and ECS — this is what lets a platform this large move quickly, and it is the prerequisite that lets the RSI loop run safely.
We build AI-native, and it is structural rather than aspirational. Our model is an agentic development lifecycle: agents own the mechanical middle — writing code, generating tests, raising CRs, deploying, monitoring — while engineers move into a steering role, choosing what to build, defining what correct means, and owning what ships. Agents do not merge their own work; the agent proposes, the engineer decides. This domain is unusually well suited to it. Benefits is rules on top of rules, and once those rules are expressed in a form a machine can read, agents can generate the configuration, write the tests that prove it, and validate a change against every country and plan we support. That is why the AI-native bet and the global expansion are the same project rather than two.
As a Software Development Engineer, you are an autonomous contributor to your team’s software. You will own components and services across their full lifecycle — design, implementation, testing, launch, operation, and documentation. On a program this new, the business problem and technical direction are set, but the implementation and the design of what you own are open, and you are expected to work them out, argue them in design review, and own the outcome. You will partner with senior engineers on the decisions that shape long-term architecture, and much of the interesting work sits at the seams between systems, where the answer must be established rather than looked up.
Key job responsibilities
• Design, build, and launch platform services and major components across the full lifecycle — working backwards from the customer through design, code, test, deployment, and operations.
• Build the configuration-driven core: model eligibility, coverage, and enrollment rules so that a new country, plan, carrier, or population is delivered through configuration and extension rather than a custom build each time.
• Build capabilities that act for hundreds of thousands of employees at once, with impact prediction, phased rollout, and safety controls that make large-scale automated action routine rather than risky.
• Contribute to building and evolving the RSI loop: design always-on agents (DevOps, product-usage, product-intake) that continuously observe the running system, propose improvements to a common backlog, and — once approved — drive those improvements through the build loop to production.
• Build simulation testing infrastructure that validates changes against realistic populations at machine speed — synthetic workers, sandboxed environments, and chaos scenarios — so the system can prove correctness across every segment we serve without a human checking each one. This is the prerequisite that unlocks autonomous iteration.
• Solve difficult distributed-systems problems in a high-volume event-driven platform: event ordering, idempotency, replay, and keeping systems in agreement as employee data changes continuously.
• Launch into new markets and new popu