# How I Scale My Startup While Sleeping Right now, as you read this, I'm probably not at my desk. I might be asleep. I might be making coffee. I might be doing something completely unrelated to software. And yet — this blog post was briefed, written, converted to HTML, and pushed to GitHub while I was doing other things. The brief? I sent it to **Claude Code** via **AgentRQ**, from my phone, in one message. That's not a party trick. That's how I've structured my entire startup — using **AgentRQ** to delegate tasks to **Claude Code** across isolated workspaces, each one running its own human-in-the-loop automation while I'm somewhere else.  ## The Problem With Being One Person Every founder knows the feeling: you have more to build, market, and sell than hours in the day. The backlog grows. The pipeline stalls. You ship one thing and three more appear behind it. The instinct is to work harder. Wake up earlier. Stay up later. But there's a ceiling on how much one human can do, and it has nothing to do with motivation. The real question is: *what can you delegate to something that doesn't sleep?* ## Three Workspaces, Three Machines Running in Parallel The key insight I learned early: you can't run marketing, sales, and product in the same context. They have different goals, different voices, different definitions of "done." Mixing them produces mediocre output across the board. So I set up three isolated **AgentRQ workspaces** — each with its own mission description that Claude Code reads before every task. They run in parallel. They don't know about each other. And they each scale themselves. But context alone doesn't scale. The real unlock is what each workspace is *scheduled to do while I'm not watching*. ## Workspace 1: Marketing The marketing workspace knows the brand voice, the target audience (Claude Code developers and solo operators), and the content strategy. It has a standing mission: surface opportunities, draft content, and keep the site growing. Here's what runs automatically: - **Every morning**: Scan HN, Reddit, and developer forums for questions about Claude Code, MCP, and AI agent tooling. If something relevant surfaces, draft a post outline and queue it as a task for my review. - **Weekly**: Review site analytics signals — which glossary terms or blog posts are getting traffic — and suggest the next three pieces to write. - **On demand**: When I send a brief (like I did for this post), the workspace picks it up, does the research, writes a draft, converts it, and opens a review task for me. When something needs a judgment call — a headline choice, a positioning decision — AgentRQ pings me. I respond from my phone. The workspace keeps going. ## Workspace 2: Cold Outreach The outreach workspace has a different mission: find developers and teams who would benefit from AgentRQ and start genuine conversations. It doesn't spam. It researches. It looks for developers writing about Claude Code frustrations, teams shipping agentic products, and open-source projects that could benefit from human-in-the-loop tooling. When it finds a strong match, it drafts a personalized message and surfaces it for my approval before anything gets sent. That last step is non-negotiable. **I approve every outreach before it goes out.** The workspace does the research and drafting; I provide the judgment and the send button. That's the human-in-the-loop part. What runs automatically: - **Daily**: Identify three new prospects, research their work, draft an outreach message - **Follow-up queue**: Flag prospects who haven't responded after five days, suggest a one-line follow-up for my review - **Weekly**: Summarize what's working — which angles are getting replies, which aren't ## Workspace 3: Coding The coding workspace is where the product gets built. It knows the codebase, the architecture decisions, the open issues, and what "done" means for each type of task. It runs a different kind of automation: - **On every commit**: Check for broken internal links, missing meta tags, and inconsistent navigation across all pages - **Weekly**: Review open issues and draft implementation plans for the top three, queue them as tasks for my review - **Nightly**: Run the full linter suite across glossary and blog sources, flag anything that fails When the coding workspace hits a decision point — "should I refactor this or ship it as-is?" — it creates an AgentRQ task and waits. I unblock it in two minutes from wherever I am. It ships. This week alone, the coding workspace propagated a nav change to 49 pages, added the blog infrastructure, and wrote two posts. ## This Post Was Written by Claude Code Here's the meta moment that makes this concrete: I wrote this blog post by sending a single task to Claude Code via AgentRQ. I sent the brief from my phone. Claude Code drafted the post in markdown, ran the converter to generate HTML, and committed everything to the repo. I reviewed it via AgentRQ reply, asked for tweaks, and approved. Total time on my end: about fifteen minutes, split across two phone sessions while I was between other things. The brief-to-published loop ran entirely through AgentRQ. You're reading the output. ## It's Not Autonomous. It's Collaborative. The thing most automation tools get wrong: they try to remove the human entirely. That works for deterministic tasks. It fails for anything that requires judgment — and most real work does. AgentRQ keeps me in the loop without making me the bottleneck. The workspaces do the research, drafting, and mechanical execution. I make the calls that actually matter: approve this outreach, pick this headline, ship this version. When I'm asleep, the workspaces queue up decisions for the morning. When I wake up, I have a set of high-leverage choices to make — not a pile of grunt work to do. That's not replacing my judgment. That's giving it more leverage. ## Getting Started If you want to set this up yourself, the path is short: 1. [Connect Claude Code to AgentRQ](/docs/getting-started) — takes about 60 seconds 2. Create one workspace per domain (marketing, outreach, coding) with a clear mission description 3. Set up your first scheduled task in each — start low-stakes, like a daily research scan or a weekly link audit 4. Define your approval gates — what decisions need your sign-off before anything ships? 5. Build from there The workspaces and cron features are available on all plans. The more precise your mission descriptions, the less you have to explain every time. The night shift is waiting.