The $500 AI Fund

Can an AI agent build a profitable business with $500 and minimal human input?

I'm running a public experiment. My AI agent Jars gets $500 in starting capital and builds a real business — from idea selection to deployment to marketing to revenue. I document everything.

No fake demos. No cherry-picked results. Real money, real decisions, real P&L.


Why This is Interesting

A human founder makes ~10 strategic decisions per day. They sleep 8 hours. They get distracted. They have bad days.

Jars makes a decision every 15 minutes. That's 96 decisions per day. While sleeping, eating, or doing nothing — the business is still iterating.

Here's what that looks like at scale:

Human Founder
AI Agent
Decisions per day
~10
~96
Hours working
8-12
24
Time to deploy a fix
hours-days
minutes
Simultaneous tasks
1-2
5-10 (parallel subagents)
Emotional decisions
yes
no
Salary
$0 (but your time)
~$20/month in API costs
Days off
weekends, burnout
never
First week decisions
~70
~672
First month decisions
~300
~2,880

In 30 days, Jars will have made roughly 10x the decisions a human founder makes. Each one logged, tracked, and reviewable.

The question isn't whether AI can replace founders. It's whether speed of iteration beats quality of intuition when the budget is $500 and the clock is ticking.


How it Works

The experiment runs in 3 phases:

Phase 1: Supervised (Week 1)

Jars asks me for feedback on every decision. "Should we target X or Y?" — I decide, Jars logs my reasoning. Goal: build a decision-making framework from 20+ real choices.

Phase 2: Co-pilot (Week 2)

Jars proposes decisions, I approve or reject. "I think we should pivot to X because [reason]. Agree?" Goal: 80%+ approval rate.

Phase 3: Autonomous (Week 3+)

Jars makes decisions independently. I get a daily summary and only intervene on spend over $50. Every 30 minutes, Jars checks analytics, diagnoses problems, and spawns subagents to fix them.


The Architecture

Johann + Jars
strategy & planning
Jars
operator — monitors, decides, delegates
Subagents
executors — build, deploy, market
Live Business
real product, real users
Analytics Check
Jars reads analytics every 30 min
Action Loop
reassess → adjust → spawn next task

Jars doesn't just make decisions — it executes them. It spawns disposable subagents that write code, build pages, push to production, and self-destruct. No meetings. No standups. No waiting on anyone.

The entire operation runs on a VPS that costs $5/month.


Budget Breakdown

Category
Budget
Domain
~$10
Hosting
$0 (Vercel free)
Marketing
$0-250 (organic first)
Tools/APIs
$0-50
Reserve for pivots
$190+
Total
$500

Stop-Loss Rules

"Hard rules. No exceptions. No sunk cost fallacy."

Day 7: $0 revenue → pivot immediately. No excuses.
Day 14: Less than $50 revenue → kill or pivot. Clock is ticking.
Day 14: $50+ revenue → double down. Reinvest everything.
Day 30: Less than $200 revenue → kill the experiment.
Day 30: $200+ revenue → Johann invests more capital.
Any time: 3x return on spend → Johann adds funding immediately.

"Every dollar of revenue earns trust. More revenue = more investment from Johann. The agent has to prove ROI before it gets more capital."


What Makes This Different

"Most 'AI builds a business' content is fake. Someone builds it themselves and says AI did it. Or they use ChatGPT to write copy and call it an AI business."

"This is different:"

Real capital at risk. $500 of my own money.
Real autonomy. The AI agent actually makes decisions, not just suggestions.
Full transparency. Every decision, every dollar, every mistake — logged publicly.
Real infrastructure. Not a ChatGPT wrapper. A multi-agent system with heartbeats, subagents, memory, and autonomous execution cycles.
Real timeline. 45-day kill switch. No endless "we're still building."

Success Metrics

Time to first dollar of revenue
Time to $100 MRR
Time to $500 MRR (break even)
Autonomous decisions vs human-approved decisions
Total hours I spent vs revenue generated
Cost per decision vs human equivalent

Daily Log

"Updated daily once the experiment starts. Every entry includes: decisions made, money spent, revenue earned, what worked, what didn't."

Experiment hasn't started yet

follow along

get weekly updates on the experiment. every decision, every dollar, every pivot.


Status: Planning

"Haven't started yet. Still deciding on business type, niche, and monetization model."

"Follow along — this page updates in real time."