SlipstreamJobs tracks this role from the company's public career site. Apply directly on the employer's site.
Clutch builds AI agents for credit unions. Emma, their collections agent, is live at five credit unions and handled 60,000+ member conversations in twelve weeks. You'll be one of two engineers on an agent pod, taking it from "sold" to "live and performing" at specific credit unions.
Your role bridges engineering and customer delivery. You sit in discovery sessions, build integrations to core banking and loan systems, author and tune agent behavior, run live tests with credit union staff, and stay through go-live and early production. You're based in the US and can reach Midwest or Southeast credit unions within a day or two when needed—roughly 20% travel on average, with one- to two-week stretches around kickoffs and go-lives.
This is a senior engineering role, not solutions or support. You write production code in the monorepo every week: integrations, modules, and agents themselves (prompts, tools, evals). You make judgment calls in front of customers and own outcomes. What you learn in the field flows back into the product as defaults, playbooks, and platform capabilities.
The Agents team is small and senior: seven engineers, a dedicated AI product group, and AI sales specialists working directly with large credit unions. Each agent is staffed by a pair of engineers—one leaning toward conversation and LLM, one toward software and integrations—with a pod lead owning the product day to day. No layer between you and the customer; no layer between you and the platform team. Operating style is direct, written, evidence-driven: BLUF updates, evals over opinions, business outcomes over ticket counts. The team spans Brazil, Argentina, and the US, working async by default.
Within 1 month: ship your first production change to a live agent, learn one agent deeply (system prompt, tool contracts, orchestration, compliance), join at least one credit union implementation as second engineer including live testing, and map where the current process loses time.
Within 3 months: own the technical side of one credit union implementation end to end (core and LOS integration, agent configuration, testing, go-live, first weeks of production), run the eval and tuning loop, complete at least two on-site visits and turn findings into product decisions, and contribute at least one reusable capability back to the platform.
Within 6 months: take one net-new agent from sold to live at its first credit union with measurable data, become the engineer credit union staff ask for by name, write the implementation playbook well enough that AI Ops runs the second and third credit union without you, and shape what the next agents look like.
Requirements:
- 5+ years of software engineering in a fast-paced environment, preferably SaaS and/or Fintech
- Senior TypeScript and Node.js backend experience in a large codebase (NestJS, CQRS, schema-validated boundaries, dependency injection)
- Integration and API engineering: secure service-to-service endpoints, third-party APIs, webhooks, legacy banking system integration
- Security and auth judgment: OAuth and machine-to-machine auth, tenant-scoped access, secrets handling, no PII in logs
- Customer presence: can sit across from a VP of Collections, hear a request, and make judgment calls in the room
- Code-quality judgment under deadline pressure: take on debt deliberately, name it, pay it down
- Willingness to travel roughly 20% on average with one- to two-week stretches around kickoffs and go-lives; based in US, able to travel on short notice
- Genuine appetite for LLM agents: eval discipline for non-deterministic systems, reading system prompts like code, understanding that some decisions must never reach the model
- Nice to have (not required): voice and telephony (LiveKit, Twilio), Temporal or event-driven orchestration, Python, core banking or LOS integration experience, credit union or banking domain knowledge
Note: Prior LLM or ML experience is not required. The company looks at what you have shipped and how you reason about it, not years of AI experience.
