AI Agent Observability Engineer

The Role

We're building the products that tell enterprises whether their AI agents can be trusted — and we need someone who works end to end, from an ambiguous problem statement through research, prototyping, and production. You'd get the problem, not the spec: research the approaches, prototype, prove what works, build it, and integrate it into the platform alongside our engineering and data science teams. This role exists because agent observability moved from roadmap to revenue faster than anyone predicted, and the work is now on the critical path.

What You'll Do

• Take an open problem end-to-end — from research and prototyping through production, killing what doesn't work before it becomes someone's roadmap

• Design and ship agent-powered features — root-cause analysis, incident triage, monitor generation — and integrate them into the platform with our engineering team

• Build the eval infrastructure that makes those features safe to change: golden datasets, regression suites, offline and online scoring, and the judgment calls about what "good" means

• Own retrieval and context pipelines over customer metadata, lineage, and query history, and instrument agent behavior in production — traces, failure taxonomies, cost and latency budgets — to close the loop on quality

• Partner with data science on detection quality and experiment design, and with PM on what an agent should do versus what it merely can do

• Set the technical bar for how we build with LLMs — patterns, guardrails, and the internal tooling other engineers reuse

What We're Looking For

• You've built agents in production. Not integrated a framework. Not worked on a team that had one. Built them — agents with real autonomy and internal loops, where the model uses tools and decides what to do next without a human in the middle, and you kept them running once real users showed up. RAG with a wrapper doesn't count. Neither does a set of MCP tools pointed at an API.

• You've run evals and monitored agents after launch. Agents are non-deterministic, so normal tests don't work on them. You've owned an eval framework — golden datasets, regression suites, offline and online scoring — not a folder of one-off scripts. And you've watched agents in production, not just in dev.

• Python, plus an ML or data science background. Python is your daily language and you're solid on the backend, though you don't need to be a distributed systems specialist. You understand models well enough to reason about how they behave — you're not an application engineer calling someone else's API.

• You work from a problem, not a spec. Handed an ambiguous problem statement, you design the experiment, build the smallest version to test it, and take what works into production.

• You use AI tools every day. Claude or its equivalents are part of how you write code and do research, not something you tried once. This is backend and model layer work, by the way — no frontend.

• You'd rather ship than polish. Most of this work needs a good answer quickly, not a perfect one eventually. You can tell which problems are the exception and deserve real depth — and you'll say no to the version that demos well and falls apart in production.

Nice to have: statistics and hypothesis testing, applied rather than theoretical. Building and maintaining MCP servers. Experience in the data and cloud space — Snowflake, Databricks, dbt, Airflow.

This Is Not For You If

• Your AI work is retrieval with a wrapper, or MCP tools pointed at an API — nothing that decides and acts on its own

• Your LLM experience is prototypes, notebooks, and demos that never carried production traffic

• You need a fully specified problem before you start, or you're uncomfortable with the ambiguity of a category being invented in real time

Why Monte Carlo

• We created the data observability category and we're doing it again with agent observability — you'll build where the market is forming, not where it's settled

• Series D, $236M raised, backed by Accel, Redpoint, Notable Capital, ICONIQ Growth, and Salesforce Ventures

• Customers include HubSpot, Fox, Nasdaq, Toast, and Mercado Libre — your work ships to enterprises with real stakes

• Snowflake Partner of the Year and a verified connector in Anthropic's Claude AI directory

• Remote-first by design since day one, and recognized as a Best Workplace for it

• Competitive compensation, equity, and a remote-first environment.