Running a one-person company on AI agents for 6 months: a 10-part framework
For six months, one person ran a company almost entirely through AI agents out of a single , covering marketing, sales, CRM, content, and outreach as actual rather than just email drafting. The framework wasn't planned in advance; it emerged from trying to eliminate admin work, and each of its ten parts includes both what worked and where it broke. The first principle is treating context as code: instead of wiring the AI into ten different SaaS tools, the entire company is moved into inside one repository, with each department as its own folder, because AI is much better at reading and writing files than clicking through .
This shift took the AI from being nearly useless across scattered to genuinely running parts of the business. Over time, though, the folders grew large and the AI's ability to recall earlier context degraded, a problem now commonly called ''.
Key points
- A one-person company ran almost entirely on AI agents for 6 months, out of a single git repo
- Covers real — marketing, sales, CRM, content, outreach — not just simple tasks like email
- Core principle: move the whole company into plain files instead of connecting AI to many SaaS apps, one folder per department
- This let the AI go from useless across to actually running parts of the business
- Downside: as folders grow, the AI's recall degrades — a problem termed ''