Artificial Intelligence Engineer

At The ReWork Group, we partner with high-growth startups and forward-thinking companies to build the future.

Most agents are demos. This one will touch money.

When an agent hallucinates in a chat app, someone gets a bad answer. When an agent hallucinates in payments, someone loses a deposit, misses a chargeback deadline, or trips a compliance review. That gap is the whole job.

We're looking for an engineer who has built agent harnesses on commercial LLMs, watched them survive months of real production traffic, and knows the difference between "it works" and "it works on the thousandth edge case at 2 a.m."

Our client is building a new product line: customer-facing agents that their platform partners embed directly into their own software, the same way they embed payments today. You'll be one of the first engineers on this team, shaping the product, the architecture, and the standards from day one.

What you'll own

• The agent harness. Build and evolve the harness behind their customer-facing and internal agentic products, taking them from working prototype to production system serving many customers.
• Domain expertise as agent capability. Translate deep payments knowledge (disputes, chargebacks, refunds, payouts, merchant onboarding, KYB/KYC) into skills, knowledge bases, and runbooks agents can rely on in specific situations.
• Policy enforcement before execution. Enforce rules at the harness level so every action an agent attempts is checked against clear, auditable policy before it runs. Think idempotency, permission scopes, spend limits, and human-in-the-loop checkpoints where money moves.
• MCP connectors. Build agentic connectors with MCP, including authentication and permissioning for agents acting on behalf of merchants and platform users.
• Security and data governance. Manage data retention, isolation, and security using AWS Bedrock AgentCore, in an environment shaped by PCI DSS, SOC 2, and the expectations of our banking partners.
• Scale before it hurts. Anticipate the failure modes that only appear at volume: drift, cost blowups, cascading tool errors, latency under load, and the long tail of weird merchant data.
• Internal agents. Build agents for teams across the company, starting with Risk, where faster, well-reasoned decisions on fraud and underwriting have direct P&L impact.
• Project leadership. Scope and lead projects within the squad and partner closely with an incoming product manager to define what we build and why.

Who you are

You've shipped agents that real users depend on, and you've been paged when they broke. You treat LLMs as powerful but unreliable components that need guardrails, evals, and observability. You like startup pace but have the judgment to slow down when money, data, or trust is on the line. You'd rather own a problem end-to-end than hand off pieces of it.

What you bring

• Production agentic experience. Hands-on experience building agentic products that run in production, not just prototypes.
• Harness depth. Experience building agent harnesses on commercial LLMs, ideally Anthropic's Claude and OpenAI models.
• Core agent design. Strong working knowledge of persistent memory, tool and skill design, and multi-step task execution.
• Tool integrations. Experience building MCP servers or similar tool integrations, including auth and permissioning for agents acting on a user's behalf.
• Seniority. 6+ years of software engineering experience, with the last 2 focused on agentic or LLM-based systems.
• IC leadership. Comfort working as an individual contributor who scopes and leads projects without direct reports.
• Education. Bachelor's degree in Computer Science or a related field, or equivalent work experience.