OpenClaw Agent Engineer - Build AI Agents and Keep Them Running

READ THIS FIRST

You must have used OpenClaw before. You must be able to prove it.

If you have not used OpenClaw, please do not apply. We will not read your message. We are saying this clearly so we do not waste your time or ours.


THE JOB IN ONE SENTENCE

You build OpenClaw agents. You keep them running. You make them faster, cheaper and more reliable every week.

That is the whole job.


ABOUT US

We are a health and wellness company. We sell products online. We have a small remote team in several countries.

We are building a team of AI agents to run parts of our business.


WHAT IS AN AGENT HERE

An agent is a small program that does one job. It runs by itself on a timer. Nobody starts it by hand. It runs at 3am while everyone is asleep.

We want 10 to 20 agents. Each one does one job.

Three real examples:

1) STORE WATCHER. Every 3 hours it checks our online shop. Is checkout working? Is any product priced wrong? Are we selling stock we do not have? If yes, it reports. If no, it says nothing.

2) INBOX READER. Twice a day it reads email, Slack, Asana and meeting notes. It works out what is waiting on the owner. Then it writes the replies for him to approve.

3) COMPETITOR WATCHER. Once a week it checks what our competitors changed. It only reports the changes.

Above all the agents sits one main agent. We call it the Chief of Staff. It manages the other agents. It is the only agent that talks to the owner.


WHAT YOU WILL DO

1) BUILD THE PLATFORM. One AWS account. OpenClaw installed and running on a timer. A safe place to store passwords and API keys. A log of every run. A stop button.

2) BUILD THE AGENTS. One at a time. We give you a written brief for each one. You d ---------- how to build it.

3) KEEP THEM RUNNING. This is half the job. An agent that stops quietly is worse than no agent at all, because we will trust a silence that means nothing is being checked.

4) MAKE THEM BETTER EVERY WEEK.
- Using too much money? Make them cheaper.
- Too slow? Make them faster.
- Sending too many messages? Reduce the noise.
- Failing sometimes? Find out why and fix it.

You will send a short report every week: how many runs, how many failures, how much it cost, and what you improved.

5) OWN IT. You will be the only person who understands this system. We need you to write things down so that stays safe.


YOU MUST HAVE

1) OPENCLAW EXPERIENCE. This is not optional. You have installed OpenClaw yourself. You have run it. You have fixed it when it broke. We need proof. See the PROOF section below.

2) AWS. You have set up EC2 yourself. You know IAM. You have used Secrets Manager or something like it.

3) PYTHON. You write code that runs on a server. Not only in a notebook.

4) API EXPERIENCE. REST APIs. OAuth. Webhooks. You have handled rate limits and retries. You have worked with an API that gave you wrong data.

5) LLM API EXPERIENCE. OpenAI, Anthropic or similar. Inside real code. Not only in a chat window.

6) CLEAR WRITTEN ENGLISH. Most of our work is written. You must be able to explain a technical decision in writing.

7) TIME OVERLAP. At least 4 hours per day that overlap with Sydney, Australia business hours.


PROOF OF OPENCLAW EXPERIENCE

Your application must include at least one of these. Two is better.

OPTION A - A SHORT SCREEN RECORDING. 2 to 3 minutes. Record your own screen. Show OpenClaw running on your machine or your server. Show the config. Show the logs. Show it doing something.

OPTION B - A REPOSITORY. A GitHub link with your commit history, showing your OpenClaw setup or an agent you built for it.

OPTION C - A LIVE SYSTEM. Something running now that we can look at, or that you can show us on a call.

Screenshots alone are not enough. Anyone can send a screenshot. We need to see you working with it.

AND ANSWER THESE FIVE QUESTIONS. These are simple if you have really used it. Please answer all five.

1) Did you install OpenClaw yourself, or did you use a one-click hosting service or a template? Both answers are fine. We just need to know which one.

2) What machine did it run on? Say the operating system and where it was hosted.

3) How did you keep it running after you closed your laptop or logged out? Name the exact method you used.

4) Where do you look when you want to see what OpenClaw did? Name the file or the command.

5) Tell us about one time it broke. What broke, and how did you fix it?


PLEASE DO NOT APPLY IF

- You have never used OpenClaw
- You cannot show us proof
- You have only used no-code tools and never written code that runs on a server
- You want a job where every task is fully explained before you start
- You are sending the same message to many jobs today


ONE MORE THING, AND WE MEAN IT

We need you to tell us the truth about what you can and cannot do.

If something is too hard for you today, say so. We will pay for the course. We will give you the time.

What we cannot work with is a date that was never real, or a yes that was really a maybe.

Before you start each piece of work, we will ask you to rate it from 1 to 10. 1 is very easy. 10 is very hard. We want the real number. If it is an 8, say 8.

This is a real part of the job. It is not a nice sentence in a job advert.


HOW TO APPLY

Send a message with all five things below. If anything is missing, we will not read it.

1) Start your message with the word HEARTBEAT. This shows us you read the post.

2) Your proof of OpenClaw experience. Recording, repository or live system.

3) Your answers to the five OpenClaw questions above.

4) Your answers to these three questions. Short answers are better than long ones.

a) Tell us about one program you built that ran by itself for 30 days or more. Something broke while nobody was watching. How did you find out it had broken?

b) Your program runs at 3am. Nobody is logged in. It needs an API key to do its job. Where do you keep that key, and why there?

c) Twenty different agents all send messages to one busy person. After one month, that person stops reading them. What do you build so this does not happen?

5) Your rate. USD per month, for 40 hours per week.

Please write your application yourself. We read a lot of AI writing every day. We are hiring for good thinking, not for good grammar.


WHAT HAPPENS NEXT

1) We read your proof and your answers.
2) A 30 minute video call.
3) A paid test project. One week. One small agent. We pay your normal rate. You keep the money even if we do not continue.
4) If it goes well, full time and long term.