The AI Org Chart I Run With Zero Headcount
I'm an EM with a side brand and no team, so I built one out of AI — a co-founder at the center and a dozen specialists, each with one job. Here's the full org chart and how to make your first hire this week.
- #agents
- #orchestration
- #claude-code
I manage engineers by day and I run a content brand on the side, and for a long time the brand had a headcount of exactly one: me. Then I stopped thinking of my AI tools as apps and started thinking of them as an org chart. Same instinct I use at work — the question isn't "which tool," it's "who owns this."
The three layers
Draw it like a real org. At the top is me, making the calls. One level down sits Claude as my co-founder and chief of staff — the one hire that knows the whole business and connects everything else. Below that: the specialists, each a different tool with one narrow job, exactly like people you'd actually hire.
There's an infrastructure strip most people skip on day one, and you should too. The custom agents themselves I put together in Claude Code, the always-on ones run on a small hosted box so they don't die when I close my laptop, and they ping me in a chat channel — I use Telegram — like a teammate dropping an update. None of that matters in week one. What matters is the co-founder — that's the one thing this whole guide builds toward, and its starter prompt sits at the bottom.
The one rule that keeps it from collapsing
One tool, one job. The way this fails is hiring a single AI to "handle everything" vaguely — same as a startup where one person owns four functions and does all four badly. Every member below has a single, nameable responsibility. That's the whole reason the team compounds instead of turning into mush.
The org chart
Co-founder — a Claude Project. The center. It knows my goals, my numbers, my deadlines, and the ways I self-sabotage. It runs my morning brief, filters new ideas against what I already committed to, keeps a log of decisions, and calls out the hole in a plan before I commit to it. Get it: claude.ai → create a Project → paste the starter prompt below.
Head of engineering — Claude Code. Where I build every custom agent, and the thing that actually shipped this website. It reads the repo, writes code, runs commands, opens PRs. Nothing else on this list has a higher ceiling. Get it: start from the Claude Code docs and work up through its permission levels.
Second opinion — Codex. OpenAI's own coding agent — I keep it running inside Claude Code through the Codex plugin. It reviews Claude's work, takes adversarial passes at a design, and picks up delegated background jobs — two engineers checking each other. Get it: /plugin marketplace add openai/codex-plugin-cc, then /codex:setup (needs a ChatGPT login or an OpenAI key).
Head of knowledge — Notion. The shared memory: my SOPs, content briefs, and product notes all live here. The Notion MCP server wires it to Claude so it can read and edit the base directly, instead of me pasting context into a chat every time. Get it: turn on the Notion connector in Claude's settings, or self-host the MCP server.
Email — Klaviyo. Runs the newsletter and the automated flows for the brand. My email agent audits what's underperforming, spots the flow that quietly broke, and drafts the next campaign — which then waits for my yes. Nothing sends on its own. Get it: a Klaviyo account connected to Claude via its API.
Ad creative — Higgsfield. Generates the image and video variations for paid ads from a text brief — a full creative suite with dozens of models. My creative agent briefs it off what's actually converting, so I test more angles in a week than I could ever film. Get it: higgsfield.ai — brief it from the ads that are already working.
Media buyer — Meta. The ad platform itself, read every morning by an agent that reports against my thresholds: what to pause, what to scale, where spend is leaking. It does the analysis; I approve the moves. Get it: pipe your ad account's numbers into Claude and ask for a daily summary measured against the limits you set.
Head of research — Perplexity. Live web research from inside Claude — market questions, what a competitor just shipped, a fact I need to check before anything goes out. Get it: the official Perplexity MCP server (needs a Perplexity API key).
Competitor intel — Apify. The scraper. It runs Actors that pull competitor posts, the comments under my own content, and whatever's breaking in the industry — all of it feeding the weekly plan. Get it: a free Apify account, then point Claude at the hosted MCP server at mcp.apify.com.
Community manager — ManyChat. Watches for keywords in comments and DMs the right lead magnet instantly, around the clock. If you've ever commented a word on one of my posts and gotten a link back in seconds, that was this. Get it: manychat.com, one keyword automation per magnet.
Design studio — Claude Design, plus Nano Banana and Gamma. Three desks under one roof: Claude Design handles pages and decks, Nano Banana generates the images, and Gamma spins up a one-pager fast. Brief goes in, draft comes out, and a human designer polishes whichever one wins. Get it: Claude Design comes with the paid Claude plans.
Voice input — Wispr Flow. This is the microphone for the whole team. It turns speech into clean text in any box I'd otherwise type into, well faster than my hands move — which means every hire on this list pulls far more context out of me per minute. Get it: Wispr Flow, then use it in every text field.
Meeting notetaker — Granola. It sits on every call and takes the notes; its MCP then hands those notes to Claude directly — so the co-founder knows what a meeting decided without me typing any of it out again. Get it: granola.ai, and switch on the Granola connector inside Claude.
Make the first hire the center
Don't build this in a weekend — mine took months, one hire at a time, and each new one was easier because the co-founder already understood the business. So start there. Build one Claude Project that holds your goals, screens your ideas, and drives your calendar. Make a Project, call it "Co-founder," and answer every field for real — the part about your weaknesses is where the value hides.
You are my AI co-founder and chief of staff. You sit at the center of
everything I run. Your job is to keep me pointed at what actually matters.
WHO I AM (fill once, revisit monthly):
- Me: [name; what I run — job, brand, projects]
- Ranked goals this quarter: [specific, with numbers and dates]
- How I work: [when I'm sharp, what I dodge, how I decide]
- Where I need cover: [e.g. "I chase shiny things", "I dodge admin"]
- Non-negotiables: [family time, hard limits, lines you never cross]
WHAT YOU KNOW (paste or attach):
- The business/projects: model, current state, key numbers
- The team: who owns what, including the AI tools above
- Live priorities and deadlines
- Decisions already made, with the reasoning, so we don't relitigate
HOW YOU OPERATE:
1. Morning brief: on request, give me today's must-dos, what's overdue,
what needs my decision, and the one thing that makes today a win.
2. Idea filter: check every new idea against my ranked goals and my
shiny-thing habit. Tell me what it would displace before I commit.
3. Memory: log every real decision. If something contradicts a past one,
flag it: "On [date] you chose X because Y — what changed?"
4. Push back: name the hole in my plan before helping me run at it.
5. Synthesis: I'll paste outputs from the other agents. Tell me what
matters across them, what conflicts, what's already handled.
RULES: one clear recommendation beats three options. Plain words.
When unsure, ask one sharp question instead of guessing.
Then grow it in this order
First week, stand up the co-founder and plug in the connectors you already live in — Gmail, Calendar, Notion. Second week, add the voice input and the notetaker, so getting context into the team stops being work. Third week, bring on one specialist: pick the role tied to wherever you make the most money or lose the most hours. From there it's a single new hire each month — and by then the co-founder can draft the instructions for every one of them with you.
Sources
Keep reading
More guides like this one.
Split Your Agents by Context, Not by Job Title
Most multi-agent setups fail because people split the work the way they would split it across a team — planner, coder, tester. Anthropic's own guidance calls that the anti-pattern. Here is the split that works, and the bill you are signing up for.
Skill, Workflow, or Agent? One Question Decides
Most devs reach for an agent when a 20-line workflow would be cheaper, faster, and deterministic. Here's the three-tier ladder with Anthropic's own definitions — and the single question that tells you which one to build.
Loop Engineering: When You Stop Prompting and Start Managing
The Head of Claude Code says he doesn't prompt anymore — loops prompt for him. Here's what a loop actually is, the pieces that make one, and a copy-paste charter to run your first without writing code.