Software Dev Engineer, Benefits Experience and Technology (BXT)

Using your benefits today means portals, phone calls, hold music, and fine print. We are changing that by bringing AI in as a strategic partner: software that understands what a person needs and helps them find the answer at the moment of need. This posting is for the tech lead on one of our two biggest product bets, driving large, ambiguous problems that span multiple organizations. This is a generative-AI and agentic-engineering role first, at very large scale.

You will join Benefits Engagement Technology and Services (BETS), the Amazon organization reinventing how more than a million employees and their families engage with their benefits. Our vision is benefits that are AI-powered, personalized, and proactive, where timely guidance reaches people when they need it rather than making them hunt for it. BETS builds and operates the platform that delivers benefits experiences across mobile, web, and an AI assistant, from self-service content creation to multi-channel campaigns to analytics, and we are now building a family of AI-native products on top of it.

Two of those products are our biggest bets, and we are building each as a focused product team inside BETS:

• A reimagined health plan where AI is a strategic partner for members instead of a portal to navigate. Rather than directories, phone trees, and hold music, it understands what someone needs and brings the right answer to them at the moment of need: the in-network doctor, the real out-of-pocket cost before they book, the claim explained in plain language, and the next step taken on their behalf. It is built on secure, HIPAA-compliant foundations to serve very large member populations, with a path to serve other organizations over time.
• An agentic workforce for benefits operations that makes running benefits for a workforce of millions dramatically easier. Managing dozens of programs, vendors, and budgets is enormous operational work; AI agents take on that heavy lifting, surfacing what needs attention, preparing the analysis and the recommended action, and handling the routine, so the people who manage benefits spend their time on decisions and strategy instead of manual operations. The system learns from every decision and keeps that knowledge with the team.

Here is what sets this organization apart: we do not just build AI products, we build them the AI Native way, and we have already proven it. A small pod rebuilt a major benefits platform and shipped it to production in 3 months, work originally scoped at 12, by restructuring how we engineer around agents. In our model, humans own judgment and agents own the mechanical middle. Engineers write the specification and set the acceptance criteria; agents (Claude Code, Kiro) write a growing share of the implementation, author tests, and resolve code-review comments; engineers direct, validate, and own what ships, and still hand-write the novel, ambiguous, and infrastructure-heavy work where models fall short. Recursive self-improvement, where the system proposes its own next version and a human approves, is the north star we are deliberately climbing toward, one validated step at a time, with humans at the judgment gates.

As a Software Development Engineer II in BETS, you will own the design, development, and delivery of the capabilities that power your team's product. You will build real distributed systems and the agentic experiences on top of them, using Java and TypeScript on AWS (Lambda, DynamoDB, and more), with Amazon Bedrock AgentCore for agent runtime and MCP for composable skills. And you will build them the AI Native way: writing specs and acceptance criteria, directing coding agents through the build loop, and proving your work against realistic customer segments with simulation testing before a human ever reviews it. A successful SDE II here works backwards from the customer, brings a strong product design sense alongside a broad technical toolkit and sound system-design instincts, has a bias for ownership, and is genuinely curious about building software with agent loops rather than back and forth prompting.

How we work, honestly: teams here are small, which means high ownership density and real scope decisions rather than a ticket queue. Some of our work carries hard, externally committed launch dates, so parts of this role are deadline-driven. Because we handle sensitive personal data, privacy and security are designed in from the start and are everyone's responsibility, not an afterthought. You own what you build, including its operations and an on-call rotation. In return you build rare, portable depth: regulated, large-scale platforms and production agentic systems, plus the judgment to direct them.

This role is for builders who want to own outcomes, not tickets, and who are comfortable working on systems that do not fully exist yet. If greenfield 0-to-1 problems energize you, if you want to help prove how a small team enters a hard domain and wins, and if you see AI as the leverage that lets a handful of engineers build what used to take dozens, you will thrive here.

Key job responsibilities

• Own capabilities end to end: design, build, test, launch, and operate features and independently deployable services, including on-call for what you ship.
• Work backwards from the customer: bring a product design perspective to what you build, understanding the need behind the spec and shaping the experience, not just implementing it.
• Turn signed-off specifications into working, tested software, contributing across packages in our open-contribution model and partnering with product experts to refine requirements.
• Work the AI Native build loop day to day: direct coding agents, review human-authored and agent-authored code to the same bar, and own the correctness of what merges.
• Implement and extend the validation that earns autonomy: simulation tests across realistic customer segments, so a change is proven correct across the populations we serve before a human reviews it.
• Integrate with regulated, real-world systems (sensitive customer data, third-party and vendor APIs) with correctness and privacy as non-negotiables.
• Help improve the system that builds the system: when a manual fix reveals a pattern, propose it as a reusable skill in our shared capabilities so agents handle it next time.