About the role
from listingSenior AI Engineer I – Agentic AI, Global Commercial Services
The Role
As a Senior AI Engineer I – Agentic AI, you will be a hands-on engineer within Amex Technology, building and evolving production-grade agentic AI systems that power intelligent customer and enterprise experiences. Today that means agents that work in real time to take the tedious part of spend management off our business customers. They read messy real-world documents, work out what belongs in the record, and propose values that already fit the customer's policy, so the user reviews instead of types.
Working closely with senior engineers, architects, Product, UX, Data Science, and Engineering teams, you will design, implement, and operate scalable, reliable, and secure AI solutions. You'll work in a small team that owns the full stack of an AI product, from the agent framework and pipelines to the evals and the production service, and you'll build across all of it.
Global Commercial Services (GCS) serves millions of business customers around the world, from mom-and-pop shops to Fortune 500 companies. We back businesses so they can do more business, with a mission to be the undisputed leader in financial and membership services — responsibly driving double-digit revenue growth. We do that by offering a diverse range of payment and cashflow tools, from a wide range of traditional card products, to working capital and supply chain financing, to new digital solutions that make it easy for our customers to manage a full range of their financial and payment needs.
Why Join Our Team
The difficult part is not making an LLM demo look convincing. It is building systems that remain useful when context is incomplete, tools fail, models change, and the outcome must be explainable. At American Express, those systems must also meet the security, reliability, privacy, and control standards of a global payments company.
In this role, you will own meaningful pieces of that problem from design through production. You will take on ambiguous work within our services, help decide where deterministic software should end and model behavior should begin, build the evaluations and operating signals that prove it, and stay accountable after launch. Your work will land in both customer-facing capabilities and the shared AI platform other engineers use.
The team also treats developer experience as a product. We build and refine model access, orchestration patterns, evaluation harnesses, observability, guardrails, and AI-assisted engineering workflows. When a tool or practice proves useful, we make it reusable so teams across Amex can move faster without lowering the engineering bar.
What You'll Do
Build production agentic AI services end to end, from event trigger through LLM reasoning to persisted, surfaced results.
Contribute to our shared agent framework: orchestration, tool use, structured generation, and observability.
Build RAG and embedding pipelines on the vector search built into our operational database.
Implement evals for new agent behaviors with the team, and use them to gate prompt changes.
Build for reliability: failure classification, idempotency, DLQ handling, and safe rollout of AI features.
Participate in technical design discussions and code reviews, and grow toward owning designs yourself.
Help onboard engineers ramping onto the AI stack.
The Stack
AI platform
TypeScript
In-house agent framework on the Vercel AI SDK and Effect
Embeddings and vector search in our operational database
Offline eval harness in Go; live-traffic evals in Datadog
Services and infrastructure
TypeScript and Go services
gRPC and tRPC APIs
Event-driven pipelines on Kafka, SQS, and Lambda
EKS on AWS, feature-flagged rollouts, infrastructure as code
What We're Looking For
4+ years building backend or distributed systems in production.
Shipped LLM-powered features and can walk through how one worked under the hood: prompts, retrieval, output handling, and what surrounded the model call.
Strong TypeScript or Go, and willingness to work across both.
Solid distributed-systems fundamentals: queues, event-driven design, failure modes.
Curiosity about what an LLM should decide versus what code should decide.
Clear communication across engineering, product, and design.
Nice to Have
Experience building or maintaining internal platform tooling or frameworks.
Experience writing LLM evals or working with LLM observability tooling.
Experience with durable execution or workflow orchestration engines such as Temporal.
AI features shipped in financial services or another regulated industry.
Vector search or embedding pipelines in production.