Ramp is building the smart infrastructure for finance teams, and they are seeking an AI Agent Operator and Architect to design, build, and operate AI agents for their Marketing team. This role involves creating autonomous AI systems that enhance marketing workflows and improve over time.
Responsibilities:
- Build AI agents from scratch that autonomously run marketing workflows: content generation, campaign development and launches, paid channel optimization, creative testing, and more
- Break down workflows into the pieces agents actually need: skills, tools, evals, guardrails, memory, and feedback loops. You'll care a lot about getting this decomposition right
- Build evaluation frameworks that measure agent quality, catch regressions, and drive improvement without someone babysitting the system
- Design self-improving loops. Agents should monitor their own outputs, learn from outcomes, and get sharper over time. The bar: it works at 3am on a Sunday and is better by Monday
- Own everything from identifying which workflows to automate, to prototyping, to production deployment, to monitoring and iteration. You are the PM and the engineer
- Stay plugged into the cutting edge of agentic AI, new model capabilities, tool-use patterns, multi-agent orchestration, MCP, evals frameworks, and bring what you learn into production fast
- Build reusable agent infrastructure, internal tooling, and documentation so the rest of the marketing team can operate and trust what you've built
Requirements:
- Build AI agents from scratch that autonomously run marketing workflows: content generation, campaign development and launches, paid channel optimization, creative testing, and more
- Break down workflows into the pieces agents actually need: skills, tools, evals, guardrails, memory, and feedback loops. You'll care a lot about getting this decomposition right
- Build evaluation frameworks that measure agent quality, catch regressions, and drive improvement without someone babysitting the system
- Design self-improving loops. Agents should monitor their own outputs, learn from outcomes, and get sharper over time. The bar: it works at 3am on a Sunday and is better by Monday
- Own everything from identifying which workflows to automate, to prototyping, to production deployment, to monitoring and iteration. You are the PM and the engineer
- Stay plugged into the cutting edge of agentic AI, new model capabilities, tool-use patterns, multi-agent orchestration, MCP, evals frameworks, and bring what you learn into production fast
- Build reusable agent infrastructure, internal tooling, and documentation so the rest of the marketing team can operate and trust what you've built
- Can mock up an MVP and then go build the thing. You scope what matters, why, and then ship it yourself. PM brain, engineer hands
- Are extremely AI-pilled. You live in the frontier. You've read the papers, run the benchmarks, broken the models. You have strong opinions on tool-use vs. code-gen agents, when to use RAG vs. fine-tuning, and how to build evals that actually matter
- Think in systems and loops and not tasks and tickets. You want to design processes where agents trigger other agents, outputs become inputs, and the whole system compounds without someone pressing 'run.'
- Have built and shipped agents in production and not something theoretical. Real systems that ran on their own, handled edge cases, and got better over time
- Are process-oriented in a way most 'AI people' aren't. You know the difference between a cool demo and a reliable agent is evals, observability, structured outputs, error handling, and constant iteration on failure modes
- Experiment fast and cheap. You validate agent architectures with quick tests before sinking time into production builds. High failure rate is fine when your feedback loops are tight
- Are fluent across the modern AI stack without being religious about any of it
- Communicate clearly, own your work completely, and don't wait to be told what to build next
- Familiarity with marketing systems like HubSpot, Segment, Amplitude, Salesforce, similar GTM platforms, or a willingness to learn
- You've built multi-agent systems with orchestration layers and inter-agent communication
- You've shipped internal tools or platforms that other teams actually adopted and relied on