After running an autonomous AI agent for a week — scanning Reddit for gigs, publishing articles, building karma, tracking leads — I realized the most valuable part wasn't the code. There was no code.
The entire system is 18 markdown files.
The agent reads a "brain" file at the start of every cycle. The brain defines:
Then it reads 8 memory files to know where it left off, picks the next action, does it, and updates all the files.
memory/
├── state.md ← "save game" — where the agent is right now
├── daily-log.md ← append-only journal (never delete entries)
├── earnings.md ← revenue tracker by strategy
├── content-log.md ← published articles with performance data
├── leads.md ← lead pipeline (FOUND → CONTACTED → CONVERTED)
├── lessons.md ← patterns extracted from failures
├── opportunities.md ← freelance gig tracker
└── experiments.md ← hypothesis testing framework
The key rule: daily-log.md
is append-only. Even if every other file gets corrupted, the journal preserves the full history. This has already saved me twice.
In 7 days with zero budget:
The system works. The revenue is still a human problem.
No database. No API. No deployment. No dependencies.
Every LLM can read markdown. Every developer can edit markdown. The agent's entire state is human-readable, version-controllable, and debuggable by opening a text file.
When something goes wrong, you don't check logs — you read state.md
.
Before the agent takes any action, it answers three questions:
If it can't answer → RED by default. This single framework prevented every serious mistake.
I've packaged the entire system — brain, memory templates, config files, quickstart guide — into a starter kit. 18 files, copy and customize in 5 minutes.
Works with Claude, GPT, or any LLM that can read files.
Week 1 of the AI agent experiment. 12 articles, 18 leads, 19 lessons, $0 earned. The infrastructure is solid. The revenue is next.