AI Agent Engineer (LLMs, OpenClaw, MCP)
What the agents must do
Every agent works for a real person on our team. Your job is to make that agent the best assistant that person has ever had.
Be fully committed to its human counterpart and help them in every way possible
Be proactive: spot what needs doing and act on it before being asked
Find tasks people do by hand and automate them, done faster and better than a human would
Keep improving over time, learning how each person works
Know when to act and when to check with its human first
What you will do
Build and run our agent platform on OpenClaw (or a similar self-hosted runtime) on AWS
Build agents with multiple jobs and tasks, managed by one lead agent
Set agents up properly from the start: memory, context, goals and clear handover to humans
Choose, set up and tune LLMs, both off the shelf (Claude, GPT, Gemini) and open-weight or customised models
Build fully custom MCP servers that connect agents to email, Slack, project tools and business systems
Keep it secure, reliable and cheap to run: secrets, permissions, logging, cost limits
Test agents properly before and after every change
You must have
Hands-on OpenClaw or similar agent runtime experience, installed and run by you, not a one-click template
A deep understanding of how to set up agents that are proactive, not just reactive chatbots
Deep knowledge of LLMs: prompting, tool calling, RAG, model selection, cost and latency trade-offs
Experience running open-weight models (Ollama, vLLM or similar) and knowing when fine-tuning is worth it
Built custom MCP servers yourself, from scratch
Strong Python or TypeScript and real AWS experience
Something you built that ran unattended for months
Good written English and at least 4 hours overlap with Sydney time
Nice to have
A fine-tuned model in production
Agents that read email or chat and draft replies for approval
Slack, WhatsApp, Microsoft 365 or Shopify APIs
How to apply
Start your message with the word HEARTBEAT and include:
Which agent runtime you installed yourself, and how you kept it running after you logged out
One agent you built that did something useful without being asked. What did it do?
The LLMs you have used in production, and one time you picked an open-weight model over a hosted one
A link to a custom MCP server or agent you built (repo, video or live demo). Screenshots don't count
Shortlisted applicants do a short call and a paid one-week trial project.