Software Development Engineer, Building Connectivity Engineering

Every Amazon data center runs on networks no customer ever sees: the corporate network that engineers and tools use to reach the fleet, the segmented networks that cameras and badge readers ride on, the network for building systems, and the isolated controls networks behind power and cooling. Building Connectivity Engineering (BCE), part of AWS Networking, owns the design and operation of those fabrics in every Amazon data center, and builds them into more than 100 new sites a year. Building Connectivity Software Systems (BCSS) is the software team inside BCE. We build the controller for these networks. It holds the intended state of every fabric as a model, computes everything a new site needs (addresses, routing identifiers, ports, cabling, device registration) before a technician plugs in the first rack, renders device configuration from the model, and drives the deployment and monitoring platforms that AWS Networking already runs. The first production release went live in September 2026 and took a site build that used to need about twenty hand-run steps down to one request. Most of the controller is still ahead of us. We are hiring Software Development Engineers to build the next releases: a versioned model of intended network state with full history and rollback, device configuration rendered from that model instead of written by hand, and the orchestration that takes a site from a plan to live traffic with one physical touch. You will design services, write the design documents for them, build them on AWS, prove them in shadow mode against real sites, and run them in production. The network engineers who hold the fabric knowledge are your partners; your code is the engine their models run on. Key job responsibilities * Design and build services in AWS (Lambda, Step Functions, DynamoDB, EventBridge, CDK), modeled with Smithy and deployed through pipelines you own. * Write the design document for what you build. Every engineer on this team writes designs; it is a core part of the job, and seniority is not the gate. * Integrate with the systems of record in AWS Networking (IP address management, device registry, routing-policy store, configuration store and release, deployment, cabling plans), and find out what each contract actually does, against the real service, before committing to a date. * Prove correctness in shadow mode: compute what the automation would do for a site, diff it against what engineers did by hand, and get to zero discrepancies before any cutover. * Own what you ship in production: metrics, alarms, runbooks, and a turn in the team's on-call rotation, which covers the new services and the ones the team inherited. * Keep fabric knowledge out of the code. New fabrics, site types and rack types arrive as models the network engineers author; your job is to make the second one cost a configuration change. * Estimate honestly: size work by how novel it is and how many systems outside the team it touches, raise partner asks early, and say what is unknown. A day in the life You pick up a card to add a new reservation type to the allocation service. Before writing code you spend an hour against the real partner API in a test account and find that the contract does not do what its wiki page says, so you write that down, re-size the card, and raise the ask with the partner team that morning instead of three weeks in. The afternoon is a design review of a teammate's state machine change, where the question you keep asking is what happens on the retry, and your own code review landing with its shadow-mode diff attached. A few times a year you are the on-call engineer for a week, and the team expects that week to be your work. About the team BCSS, formed in May 2026. We bring software engineering practice to a problem space that has not had software developers, and we never cut corners: slow is smooth, and smooth is fast. Our stance in one line is a team tenet: we don't just make a change, we make a changer. Our software runs on AWS and operates entirely detached from production data traffic. We work inside an operating model where network fabric teams run like software teams: they own their business logic, expose APIs, and measure outcomes. BCSS builds the software; every fabric team owns its automation roadmap and its adoption. Design reviews are open and argued; the best objection in the room wins, whoever raised it.