AI Agent Engineer
You build the agents that run our clients' operations, from tool definitions to production monitoring.
About Hopper Labs
We are a Chilean AI consultancy. We design, build, and maintain bespoke operational agents for mid-market and enterprise companies. We work as an end-to-end operating partner: scoping, build, integration, and support, with no dev team required on the client side. The commitment is measurable ROI in under 90 days.
About the role
This is the core of what we sell. You design and build agents that do real work inside large companies' operations: they decide, they execute, they escalate to a human when they should. This is not research or demos. It is product engineering with LLMs inside, in production, with client money on the line.
- 100% remote. Where you are does not matter.
- We work against objectives, not hours. You decide when you work.
- We hire builders: autonomous people with taste, focused on quality and on adding real value.
- Real autonomy: you get a business problem, not a ticket with the solution already written.
- Technology and AI genuinely move you: you try the newest thing as it lands, you follow how the market is adapting to these tools, and you want to work out how to put them to use. We look for curious people who are deep in the industry.
What this role works with
Responsibilities
- Design and build LLM agents: tool definitions, system prompts, context control, failure handling, and human escalation.
- Own the agent architecture: when one agent is right, when several are, and when none of it is and plain code wins.
- Integrate agents with the client's real systems: ERPs, CRMs, WhatsApp, email, databases.
- Build memory and retrieval (RAG) that actually improves answers.
- Build evals: measure whether the agent got better or worse with each change instead of guessing.
- Instrument cost per token, latency, and error rate, and iterate on that evidence.
Requirements
- 3+ years building software.
- Deep command of LLMs applied to real cases: tool calling, context windows, RAG, evals, hallucination control, cost, and latency.
- You have put at least one LLM system in front of real users and kept it alive afterwards.
- The judgement to know when an agent is not the right answer.
- You track the state of the art: models, tools, and patterns change monthly and you follow it closely.
- The stack is learnable. We work with TypeScript, Next.js, and Postgres, but your technical judgement matters far more than prior experience with those specific tools.
What we look for
- A real builder: you ship to production, you do not hand over prototypes.
- Technical taste and a focus on quality: the agent has to survive the day everything breaks.
- Autonomy: you get a business problem and come back with something working.
- Obsession with real, measurable value delivered, not with a pretty demo.
How the process works
Two weeks end to end, at most.