Senior Lead Agentic Developer

About Integrant

For 34 years, Integrant has designed and delivered advanced, business solutions for U.S. enterprises in biotech, medical devices, manufacturing, finance, and related highly regulated industries. Our teams are experts in traditional web and mobile applications, AI/HPC infrastructure support, data engineering, predictive AI, machine learning, and Agentic solutions. We build long-term partnerships—two-thirds of our customers choosing to partner with us 5+ years—by solving mission-critical problems with rigor, empathy, and measurable outcomes.

Role overview

Title: Agentic Developer (Enterprise Agentic Systems)

Mission: Build production-grade AI agents that run real business processes inside enterprise environments — reliably, securely, observably, and with measurable ROI.

This is not a prompt-engineering or demo-building role. The person we hire owns the full lifecycle of an agent: design, implementation, evaluation, deployment, monitoring, and iteration. They treat an agent as a distributed software system with a probabilistic component, not as a chatbot with tools attached.

Stack-agnostic by design. We use Python and TypeScript, and frameworks such as LangGraph, the OpenAI Agents SDK, Bedrock Agents, and Copilot Studio appear in our work as examples, not requirements. We hire for the underlying engineering judgment: someone who can pick up any orchestration framework in a week because they understand what every framework is abstracting.

Key responsibilities:

1. Agent design & orchestration

• Design single-agent and multi-agent systems: planner/executor, supervisor/worker, hierarchical, and event-driven patterns, choosing the simplest pattern that meets the requirement.
• Model agent workflows as explicit state machines or graphs with defined states, transitions, retries, timeouts, and human-in-the-loop checkpoints.
• Define tool contracts (schemas, idempotency, error semantics) and implement tool-calling with validation and graceful degradation.
• Implement memory strategies: short-term context management, long-term memory stores, and session persistence with clear retention rules.
  

2. Retrieval & knowledge (RAG at scale)

• Build and tune retrieval pipelines: chunking, embedding, hybrid search, re-ranking, metadata filtering, and query rewriting.
• Handle enterprise document realities: permissions-aware retrieval, multi-tenant isolation, freshness and sync from source systems.
• Measure retrieval quality (recall, precision, groundedness) and fix it at the pipeline level, not the prompt level.

3. Evaluation, testing & quality

• Build eval suites before shipping: golden datasets, task-level success metrics, LLM-as-judge with calibration, regression tests on every change.
• Write unit and integration tests for tools, state transitions, and guardrails; test the non-LLM 80% like any other software.
• Run offline and online evals; use A/B or shadow deployments to validate changes against production traffic.

4. Observability & operations

• Instrument agents end-to-end: traces per run, token/cost accounting, latency per step, tool-call outcomes, and failure classification.
• Define SLOs for agents (success rate, p95 latency, cost per task) and build dashboards and alerts against them.
• Operate agents in production: on-call participation, incident analysis, root-cause on model and non-model failures.

5. Security, safety & governance

• Implement defenses against prompt injection, data exfiltration, and tool misuse; enforce least-privilege for every tool and integration.
• Apply input/output guardrails, PII handling, content policies, and audit logging that satisfies enterprise compliance.
• Design human-approval gates for high-impact actions and make agent decisions explainable to auditors and business owners.

6. Enterprise integration

• Integrate agents with ERP, CRM, ITSM, document management, and identity systems via APIs, event streams, and MCP-style tool servers.
• Handle authentication and authorization on behalf of users (OAuth, service principals, delegated access) correctly.
• Build agents that survive real data: inconsistent records, partial failures, rate limits, and legacy interfaces.

7. Delivery & collaboration

• Translate business processes into agent specifications with clear success criteria and ROI measures.
• Produce architecture decision records, runbooks, and handover documentation.
• Mentor engineers transitioning into agentic development; review designs and code for agent-specific pitfalls.