AI Engineer

You turn business requirements into working agent systems: composing agents, designing the prompts and behaviors that drive them, wiring in the tools and data sources they need, choosing how the agents coordinate, and deciding where a human needs to stay in the loop. Experience building or working with agentic or LLM-driven systems in a production or nearproduction setting — where a model drives control flow through tool use, multi-step reasoning, or agent-to-agent handoffs.

Overview:

You will design and build agentic workflows — multi-step, multi-agent systems that reason, use tools, call external services, and recover from failure — and take them from prototype to reliable production. You turn business requirements into working agent systems: composing agents, designing the prompts and behaviors that drive them, wiring in the tools and data sources they need, choosing how the agents coordinate, and deciding where a human needs to stay in the loop. You build on top of our agent development tooling, so much of the surrounding machinery — identity, guardrails, logging, and monitoring — is handled for you; your focus is assembling those pieces into something that solves a real problem and behaves predictably even when the underlying models do not. The work is hands-on and product-facing: you translate requirements into designs, get them working against real data, harden them for production, and keep improving them once they are live.

Responsibilities:

Design and build agentic workflows — multi-step, multi-agent systems that reason, call tools, and recover from failure — owning them from design through production.

Compose agents into reliable pipelines that integrate with external APIs, internal services, and data sources, with clear error handling and retry semantics.

Design and iterate on the prompts and behaviors that drive each agent, treating them as versioned, tested artifacts rather than inline strings, and validating them against evaluation sets.

Build the tools and integrations agents rely on — well-typed function interfaces or service endpoints — with attention to safety and correctness.

Choose the right coordination approach for each task — a supervisor directing workers, a planner and executor, competing or critiquing agents, or a fixed sequence — and know when each fits.

Decide where human review and approval belong, and build those checkpoints into longrunning workflows.

Wire in state and memory appropriately — short-term working context and longer-term memory — so agents have what they need without leaking data across users or tenants.

Build evaluation suites and regression checks that measure whether a workflow actually does its job — trajectory and outcome quality, not just pass/fail.

Harden workflows for production — diagnosing timeouts, partial failures, unexpected model behavior, and state drift in long-running, non-deterministic systems.

Instrument and debug agent behavior using tracing, logs, and cost and latency telemetry.

Contribute to shared patterns, conventions, and reusable components that raise the quality bar for how agentic systems are built across the team.

Partner with product and the broader engineering team to translate business requirements into agent designs that are practical to build, operate, and govern.

Qualifications:

Required qualifications

Strong software engineering fundamentals with production-grade Python — clean, tested code; comfortable with async patterns and long-running, stateful processes.

Experience building or working with agentic or LLM-driven systems in a production or nearproduction setting — where a model drives control flow through tool use, multi-step reasoning, or agent-to-agent handoffs.

Has composed multi-step or multi-agent workflows — orchestrating agents, managing shared state, chaining tool calls, and handling failures and retries.

Practical prompt-engineering skill — can design prompts and system messages that make model behavior consistent, and has improved them against evaluations rather than by feel.

Experience integrating with external APIs and services — REST, authentication, error handling, and designing for partial failure.

Has built tools or integrations that a model or agent calls — function interfaces, schema design, or service endpoints.

Disciplined about evaluation — can describe an eval suite they built for an agentic or LLM system, how they created the ground truth, and what it proved.

Solid debugging instincts for distributed and non-deterministic systems — can reason through failures that span multiple agents or services.

Writes meaningful tests for complex, stateful systems — unit, integration, and scenario-level coverage.

Familiarity with agent or LLM observability — tracing, logging, or monitoring of model and tool calls.

Background in backend engineering — APIs, queues, databases — that grounds you when workflows touch infrastructure.

Exposure to multi-tenant systems and data-isolation concerns. Preferred qualifications

Experience with durable or long-running workflow execution — systems where state must survive restarts, retries, or partial failures.

Familiarity with agent memory — working memory, long-term memory, and managing context windows.

Has worked with retrieval-augmented generation and reranking — wiring retrieval into an agent workflow and measuring its effect.

Exposure to human-in-the-loop patterns — approval steps, escalation policies, and review workflows in production.

Experience with agent orchestration frameworks or patterns (LangGraph or comparable).

Familiarity with prompt-injection and adversarial risks in agent and retrieval systems, and how to mitigate them.

Comfortable working against an SDK or framework abstraction — understands its constraints and when to escalate gaps.

Open-source contributions to agent frameworks or related tooling.