Job Description & Details
This is a heavy-hitting role for someone who has moved past basic prompt engineering and is actually building autonomous, production-grade agentic systems. They are explicitly filtering out junior talent here, looking instead for a seasoned architect who understands the operational pain points of running LLMs at scale. If you enjoy solving gnarly problems around state management, latency, and context degradation, this is a solid gig.
What You'll Actually Be Doing
Your day-to-day will revolve around architecting complex multi-agent workflows and building robust systems where LLMs can reliably use tools via MCP (Model Context Protocol). You won't just be writing scripts; you'll be engineering how these agents manage their memory, optimize token usage to keep inference costs down, and gracefully recover when an agent goes off the rails. Expect to spend a lot of time integrating these workflows into solid CI/CD pipelines, treating your AI infrastructure with the same engineering rigor as traditional backend services.
The Core Tech Stack
Python is non-negotiable here, and you need to be comfortable using it to bend agent frameworks like LangChain and LangGraph to your will. The real separator for this team is your deep understanding of context engineering, custom memory architectures, and tool-use integration. If you also have experience wrangling massive datasets in Snowflake or Databricks to feed your agents, you are going to be in a prime position to crush this role.
Interview Expectations
Expect them to grill you on how you handle state management and context window limits in a multi-agent environment without exploding your token costs. They'll likely ask you to whiteboard a scenario where an agent needs to dynamically decide between multiple tool outputs over several turns. The hiring manager isn't looking for textbook answers here—they want to hear war stories about how you've debugged hallucination loops and optimized token payloads in production.
Application Advice
Make sure your resume doesn't just list frameworks, but specifically highlights architectural achievements where you managed agent memory, reduced token consumption, or built custom tool-use integrations from scratch. Weave in heavy-hitter keywords like LangGraph, MCP, context engineering, and CI/CD for LLMs directly into your project summaries so you easily clear the ATS and catch the engineering lead's eye.