By the Numbers
- $50
- public framing for a monthly tool budget, not a confidential P&L
Listen
Audio placeholder · 11 min
No file in v0. The bar at the bottom is a stub. Feeds live on Listen when they exist.
The Brief
Forward this. Then decide if the essay is worth your time.
- 01
0:00 · Charter over lore
Name the roles, the inputs, the definition of done. Personality prompts are not an org chart.
- 02
0:04 · Runbooks, not mascots
Each role is a repeatable procedure with a human gate at send, spend, and publish.
- 03
0:07 · The stop rule
When does the company halt? If you cannot say, you built a chatbot with invoices.
- 04
0:10 · Public numbers only
This essay uses a public $50 frame. No employer spend. No private stack tour.
Essay
You can rent a small AI company for roughly the cost of a few streaming subscriptions. The scarce asset is not another agent persona. It is a charter: roles as runbooks, clear send rights, and a human who owns outcomes. Build the company as procedures. Do not collect digital dolls.
People ask how to "build an AI team." They usually mean a cast of chatbots with job titles. I tried the cast. It looked busy and coordinated poorly.
What worked was duller: departments as capability bundles, routed by an orchestrator, with intelligence in the coordination layer. Dumb workers. Ruthless routing. Roles written as runbooks.
Call it a $50 AI company as a public metaphor. Exact spend moves with models and tools. The point is the operating model, not a price promise.
What the company is
Think eight capability lanes, not eight persistent characters:
- Operations (deploy, fix, health)
- Intelligence (research, dig, synthesise)
- Finance (bills, forecasts, checks)
- Communications (drafts, briefs, calendar hygiene)
- Creative (media, story, assets)
- Strategy (red team, first principles, council)
- Security (guards, reviews, least privilege)
- Growth / coaching (goals, reviews, multipliers)
In my personal harness, those lanes are skill bundles triggered by language, not employees with backstories. Research keeps converging on the same warning: multi-agent theatre degrades sequential reasoning and piles coordination failure on top. Persistent agent identity is usually a tax.
So the $50 company charter reads:
- Human CEO owns send and money.
- Orchestrator routes; workers do not hold private agendas.
- Each role is a runbook: outcome, inputs, tools, must-not, definition of done.
- Draft by default. Publishing is a human click.
- Memory lives in files you control, not in a vendor thread you cannot export.
Runbooks beat resumes
A role card I actually reuse:
- Name: Morning Communications
- Outcome: One brief I can act on in ten minutes
- Inputs: Calendar sync, open loops, training plan, finance flags (personal scope)
- Must not: Invent meetings, leak private journal domains into shareable digests, auto-send
- Done: Email or dashboard with decisions and load, not a compliment sandwich
Another:
- Name: Creative Pipeline
- Outcome: Captioned short proof film from a folder
- Inputs: Interview clip, screen proof, house style
- Must not: Fake customer faces, hide consent gaps
- Done: 60-90s master + captions + human gate checklist
RivaFlow, photography, and BJJ training logs all fit this pattern. The "employee" is the procedure. The model is rented labour.
How to stand one up in a weekend (SMB)
Saturday morning: List the five jobs you already pay humans or yourself to do weekly.
Saturday afternoon: Write runbooks for two of them. Five lines each.
Sunday: Wire one orchestrator (Claude Code, Cursor, a single MCP host, whatever you already use). Point it at the runbooks. Turn off auto-send everywhere.
Monday: Run one job end-to-end. Edit the runbook where it missed. Do not add a second agent costume.
Budget mindset (illustrative, AUD, personal tooling): model subscription + light media tools + storage. Keep it near "coffee and streaming," not "stealth startup burn." If the company needs a program manager for the agents, you built theatre.
Enterprise translation
Enterprises already have departments. The failure mode is bolting twenty copilots onto twenty teams with no charter.
Translate the $50 company like this:
- Role runbooks become approved skill cards with data boundaries
- Orchestrator becomes the platform routing and policy engine
- Draft-not-send becomes workflow gates with audit
- Replaceable head becomes vendor-agnostic memory and eval harnesses
- Council / red team becomes a required adversarial pass on high-impact drafts
Procurement should ask for the charter, the eval suite, and the send policy, not the mascot names.
Failure modes
- Agent soap operas (backstories, rivalries, manager chains)
- No single human owner for outbound
- Memory trapped in chat products
- Expanding spend before one runbook works twice
- Confusing a Friday multi-agent demo with peak reliability on the second Tuesday
Closing
The $50 AI company is a discipline: buy cognition by the hour, store judgment in runbooks, keep humans on the send key.
If you want a bigger team than your headcount, do not hire digital dolls. Write the roles. Route the work. Own the outcomes.
Related
Want the next edition? Join the newsletter placeholder. No modal. No auto-send.
Ruby Wolff
Gem Alpha. Shipping in public. Builder notes, not a news desk.