Software Development Engineer, Support Agent Intelligence and Evaluation

Overview

We are building the next generation of AI-powered advertising intelligence at Amazon Ads. Our team is developing an agent platform that gives every advertiser, regardless of size or sophistication, access We are building the next generation of AI-powered advertising intelligence at Amazon Ads. Our team is developing an agent platform that gives every advertiser, regardless of size or sophistication, access to the cal

What you'll do

• We are building the next generation of AI-powered advertising intelligence at Amazon Ads.
• Our team is developing an agent platform that gives every advertiser, regardless of size or sophistication, access We are building the next generation of AI-powered advertising intelligence at Amazon Ads.
• Our team is developing an agent platform that gives every advertiser, regardless of size or sophistication, access to the caliber of strategic campaign management previously available only through dedicated agencies.
• You will work on agent orchestration, reinforcement learning from expert feedback, trust and permission systems, and scalable inference infrastructure that powers intelligent advertising decisions across Amazon's ad products.
• This is a greenfield opportunity to define how AI agents operate in one of the world's largest advertising ecosystems.
• Key job responsibilities - Develop orchestration systems that coordinate multiple specialized AI sub-agents to execute complex, multi-step advertising workflows - Collaborate with cross-functional teams (product, science, partner engineering) to define integration contracts and onboard new consumers to the platform - Build identity, authorization, and permission frameworks that enforce scoped access for diverse caller types - Implement feedback signal pipelines that capture expert judgment and route it into model improvement loops - Design and operate end-to-end evaluation systems that measure recommendation quality, detect regressions, and enforce quality gates - Partner with Applied Scientists to integrate reinforcement learning from human expert feedback into production agent systems - Own system reliability, observability, and performance for always-on agent infrastructure serving thousands of concurrent advertiser accounts A day in the life You might start the morning reviewing eval results from a newly onboarded agent skill, diagnosing why routing accuracy dropped 2%.
• By mid-morning, you're pairing with an Applied Scientist to tune the feedback pipeline that captures expert signals.
• After lunch, you're designing the permission model for a new integration.
• You end the day shipping an observability improvement that surfaces latency bottlenecks across the orchestration chain.
• Every day is different, but the thread is the same: making AI agents smarter, safer, and more reliable at scale.

Requirements

• Practical experience with Rust, Java, C++, C#.
• Experience level: 3+ yrs.
• Strong written and verbal communication in English.
• Comfortable working on-site in Seattle, Washington, USA.

Skills