Brad Mills / POW Media / Citadel OS
This is not a traditional AI engineering job. I've built a sophisticated AI infrastructure that handles most of my operations. I need someone to own it — keep it running, evolve it, and make sure it never breaks.
This is not an exploration role. The problems are already defined. The infrastructure exists. I've spent significant time and resources building an AI agent system that runs 24/7 on a Mac Studio here at the Citadel — handling daily ops, research pipelines, Telegram coordination, knowledge base management, scheduled workflows, and more.
You're not inheriting a clean system. You're inheriting a sophisticated one that needs a dedicated operator to own it.
A cron job failed overnight. You notice before I do. You diagnose the broken skill, patch it, test it, push the fix, and drop me a three-line Telegram summary of what broke and why it won't happen again. I don't know it happened until you tell me it's already fixed.
I want to automatically pull Bitcoin market signals into my morning briefing and flag anything above a custom threshold. You scope it in an hour, build it in two, test it, document it, and have it running before the weekend. I gave you one sentence. You shipped a feature.
Nothing. Because the system is running. That's the goal.
Day-to-day responsibilities
This isn't a vague "work on AI stuff" role. Here are the specific systems you'll be responsible for from day one.
The core platform everything runs on. Skills-based architecture in Node.js/TypeScript. You'll write, debug, and maintain modular skill files that give the AI agent specific capabilities. When skills break, you fix them. When new capabilities are needed, you build them.
My AI executive assistant runs on OpenClaw with semantic memory, DAG-based context compaction via lossless-claw, and persistent identity across sessions. This isn't a stateless chatbot — it's a system with institutional memory that I depend on daily. You keep it sharp, healthy, and current.
A structured knowledge base built from 240+ hours of coaching content. This powers strategic decision-making and contextual guidance inside the agent system. You maintain the RAG pipeline, keep embeddings current, and ensure retrieval quality stays high.
My media company runs automation workflows on top of the agent stack. Content pipelines, publishing integrations, research automation, and distribution tooling. You build and maintain the specific skills that power POW Media's operations.
Tailscale for private networking, 1Password CLI for secrets, Ollama for local model inference, SearXNG for private search, Alby Hub for Lightning payments. The whole sovereign stack. You maintain it, expand it, and evaluate what earns its place.
Not an exploration role. I'm not hiring someone to figure out whether AI could help my business. That question is already answered. The infrastructure exists. Your job is to operate and extend it — not to start the conversation.
Not a research position. You won't be writing papers, evaluating theoretical architectures, or running experiments to understand LLM capabilities. You'll be shipping things that work in production.
Not a "help me understand AI" role. I already understand enough to know what I need built. I need someone who can build it, not someone who starts every conversation with an explanation of how large language models work.
Not a managed position. There are no daily standups. No sprint reviews. No tickets assigned to you. You'll have clear outcomes and high autonomy. If you need a manager to tell you what to do next, this will be uncomfortable.
Not a traditional software engineering role. You won't be working on a clean greenfield codebase with perfect specs. You'll be inside a living system that evolved fast. You need to get oriented quickly, understand how things actually work, and operate confidently in imperfect conditions.
The upside of all that: Real ownership. A production system that actually matters. Work that directly impacts how a Bitcoin-native family office operates. And if you're excellent — room to grow into something more significant than a contract role.
I hire the soul, train for the role. I'd rather have someone who has shipped real AI systems with 80% of the required skills than someone with a perfect résumé who's never debugged a failed webhook at 2am. Demonstrated work beats credentials every time.
Technical requirements
I'd rather lose a good candidate by being transparent than hire the wrong person and have them inside my infrastructure. So here's exactly what gets you rejected:
Interest is table stakes. I need results. If the most impressive AI thing you've built is a ChatGPT wrapper with three stars on GitHub, this isn't the right role yet. Come back when you've got something real running in production.
This is a lean, high-trust role. If you've only operated inside large orgs with dedicated DevOps, QA, and security teams, the autonomy and context-switching here will be uncomfortable. You need to be a one-person ops department.
If I ask you to explain what a skill file does and you need 10 minutes and a diagram, that's a problem. I need someone who can give me the three-sentence version. Always.
We don't explore here. We scope, build, ship, and measure. If your default response to a new request is "interesting — I'd need to research this" before you've even tried, we're not a fit.
I can't verify you're real or capable without evidence. Given that 50%+ of developer applicants right now are AI-generated or resume-inflated, you need something concrete I can look at. A live tool, a GitHub profile, a project with actual commits — something.
This role sits inside my digital life. If you're not comfortable handling sensitive systems responsibly — or if you need a lot of reassurance about what you're allowed to do — this is the wrong fit. I need someone who treats access as a responsibility, not a perk.
My world moves fast and changes direction. If you can't operate without a perfect spec, this will be frustrating for both of us. I give you outcomes. You figure out the path.
Green flags that get you to the front of the line:
You lead with "what's the actual problem?" before talking about technology. You ask one clarifying question when you need it, then go build.
Not built once and forgot. Maintained. Debugged. Improved over time. You know what it feels like to own a live system.
"Here's a thing I noticed, here's what I think caused it, here's what I recommend, here's what it'll take." Not: "hey did you see that thing?"
You have a point of view. You can tell me why you'd choose one approach over another, and you can update that view when new information comes in. Opinionated but not precious.
I can't work with you till I work with you. This process is designed to find out what you're actually like before we commit to each other. It's fast if you're the right person. It has exit points for everyone else.
We read your application. We look at your GitHub. We check if you followed the instructions. 90% of applicants are filtered here — not by resume, but by evidence of real work and ability to follow a specific process. You either have something to show or you don't.
Timeline: 48–72 hrsThree specific questions to answer on video. 3–5 minutes total. We're not looking for polish — we're looking for how you think out loud. Can you explain a technical concept clearly? Can you be honest about something you didn't know how to do and how you solved it?
Timeline: You have 48 hrs to submitQuick call with our ops team. Culture fit, availability, logistics, basic technical orientation. We're confirming you're who you said you are and that the basics work: time zone, communication style, rate expectations. No trick questions.
Timeline: Within 1 week of video submissionWe'll send you a technical brief: a description of a real problem from the OpenClaw stack with relevant code snippets and context. You write back with your diagnosis and proposed approach. Async, no time limit — should take 30–60 minutes. We're evaluating: do you understand the problem, is your approach sound, can you communicate it clearly?
Timeline: You have 72 hrs to respondA deeper conversation about your background, how you think, and what you've actually built. We'll go deep on one or two specific projects from your portfolio. Be ready to walk through your work in detail: what broke, what you learned, how you'd do it differently. No generic answers accepted.
Timeline: Within 1 week of technical reviewThis is the real filter. A real task from the actual OpenClaw system. $400 USD. 6 hours. Sandboxed environment. We evaluate: did it work, how clean is the code, did you document it, did you show initiative beyond the spec? Full details in the next section.
Timeline: Scheduled within 2 weeks, you pick the 6-hr windowIf you've made it here, you're close. Short direct conversation with Brad — not a formal interview, more of a "let's make sure this feels right on both sides" call. Brad will tell you what he's building and what working together actually looks like. You should have questions.
Timeline: Within 1 week of test project reviewContract terms, rate, scope, communication protocols, access provisioning. We start with a 30-day engagement with clear deliverables. At 30 days we review and either expand or part ways. No ambiguity about expectations going in.
Timeline: Rolling start date, goal is <2 weeks from offer⏱️ Total timeline target: 4–6 weeks from application to start. We move fast when the person is right. If you don't hear back within the stated window for any stage, you can follow up once. After that, assume it's a no.
This is Stage 6 of the hiring process. It's paid. It's real. And it tells us everything we need to know.
You'll receive access to a sandboxed OpenClaw environment. Inside it, you'll find a skill file that is partially broken. The skill is designed to pull data from an external API, process it with an LLM prompt, and deliver a formatted summary to a Telegram channel via the agent system. It's failing in production — we'll tell you that much. The rest is yours to figure out.
SKILL.md + implementation) with no additional context$75–$120 USD/hour depending on experience and what you demonstrate in the process. This is not a race to the bottom. If you're the right person, I'd rather pay fairly and get the best version of your work than underpay and get half of it.
We start at ~15–25 hours per week on retainer. Async-first. You pick your working hours as long as critical issues get same-day response and urgent issues get sub-2-hour response. As trust builds and scope grows, hours can expand.
Initial 30-day contract with defined deliverables. After 30 days: mutual review, then month-to-month retainer if we both want to continue. Either party can exit with 2 weeks notice after the initial period. No lock-ins. If you're delivering, you'll know it.
This role has a real ceiling — in the good direction. As the system grows, as the family office expands, as more integrations get built, the scope of this role grows with it. There's no artificial cap on what this becomes if you're excellent. Some people in roles like this become indispensable partners, not contractors. That's possible here.
No equity at this stage. No benefits (it's a contract role). No guaranteed hours above the retainer floor. No hand-holding. If those things matter to you, this isn't the right fit right now.
Direct access to a genuinely interesting and sophisticated AI system. Real ownership of a production environment. Work that matters. A principal who respects your time, pays on time, and won't make you sit in status meetings. And if you're excellent — real room to grow.
The application is itself a filter. If you can't follow these instructions carefully, you won't make it past Stage 1. This isn't bureaucracy — it's a real signal of whether you read things carefully and do exactly what's asked. Every detail below is intentional.
Send your application to team@bradmills.ca with the subject line: Alfred Application — [Your First Name] — [One sentence about the most interesting AI thing you've built]. If the subject line doesn't match this format, it goes to the bin automatically.
No slides. No screen share. Just you, talking to camera. Include the link in your email.
Cover these three things:
1. Walk me through the most complex AI system you've built or maintained — what it does, how it works, and what broke along the way.
2. Why this role specifically — not "I love AI," but what about this setup makes you want in.
3. Show me one thing you've built. Pull it up, walk me through it. 60 seconds max on the demo.
Loom, YouTube (unlisted), Google Drive — I don't care how you host it. If you can't figure out how to record and share a short video, you can't operate AI infrastructure. No video = automatic rejection.
1. Two paragraphs max on why you're the right person for this role. Not your life story. Not a cover letter. Two paragraphs. What have you built, and why does this role fit?
2. One link to your best piece of AI work — a GitHub repo, a live tool, a write-up with code, something concrete. If you don't have anything to link, apply again when you do.
3. Your current hourly rate and available hours per week.
At the bottom of your email, after everything else, answer this in 2–3 sentences: "A cron job that sends a daily Bitcoin price summary to a Telegram channel has been silently failing for 3 days. The bot shows as online. Where do you look first?" If you skip this, your application is automatically rejected regardless of everything else.
No attachments (no PDFs, no resumes). No "I'm a fast learner" language. No "I'm passionate about AI." No "I'm happy to learn whatever stack you use." Tell me what you've built, show me evidence, answer the question. That's the whole application.
We review applications on a rolling basis. Strong candidates move fast.
📬 team@bradmills.ca
No recruiters. No agencies. Real humans only.
The AI infrastructure I've built handles things most people haven't thought to automate yet. It's genuinely sophisticated — and it needs a human who can keep it running, push it forward, and make it better. That human needs to be excellent, trustworthy, and technically sharp. If that's actually you — not the resume version of you — I want to hear from you.