The AI system I use to run my business alone

AI Running Solo

My AI operating system is the toolkit and agents I built to handle repetitive work instead of hiring people. I work alone with no assistant or operations team, but every day I still have dozens of small tasks: note-taking, replying to messages, saving documents, monitoring servers, writing content, downloading and processing videos. This post describes how that system is running as of October 2026, using a Mac, a Windows machine, n8n, Obsidian and Telegram, so you can see how one person working alone can organize AI. This is my own way of doing things, not a general recommendation yet. For example, you don't need two machines to start: a small n8n workflow is enough. The post has four parts: role division, automated processes, three-layer memory and lessons learned.

How does the role-division system work like a small team?

My system is a small team of three "members", each part doing one clear job instead of using one chatbot for everything. The way I operate as of October 2026, the three parts are:

• Coordinating brain running on a Mac: where data lives, automated workflows and gateways to AI models.
• Executive arm running on a Windows machine: an AI agent that takes jobs from the brain, operates the browser, downloads and extracts video, replies via Telegram.
• Knowledge vault is an Obsidian vault: strategy, processes, daily logs are all markdown files that both I and agents can read.

The two machines connect through a private network, so the agent on the Windows machine can call services on the Mac without opening ports to the Internet. For example, when I send a video download request via Telegram, the executive arm takes the job, downloads it, then the brain records the result in the knowledge vault. In the end, I just open the vault to review.

How do automated processes and AI gateways work?

Most repetitive work goes through n8n, a drag-and-drop automation tool I install myself. Each workflow is a chain of steps: receive data, process, call AI if needed, then record results in the right place. Agents also send jobs to each other through a webhook on n8n. Each message has the same four-part structure: who sends, who receives, job type (assign work, return results, alert, update status) and content. For example, a workflow receives a message via Telegram, calls a model through 9Router, then records the result in Obsidian. Thanks to this common structure, I can add a new agent without rewriting how they talk to each other.

Every AI call goes through a routing gateway (I use 9Router), so I can swap models for each task without fixing each workflow. First, work that needs careful reasoning uses a strong model. Next, simple work uses a cheap model. This way I'm not locked into a single provider.

How does three-layer memory work?

Three-layer memory is how I overcome the biggest weakness of chatbots: forgetting. The three layers are:

1. Short-term memory for events, habits, commonly used parameters. Agents write and read back quickly.
2. Knowledge database (Postgres with semantic search) for long documents, scripts, large knowledge. Documents are cut into chunks and converted to vectors so AI finds the right chunk.
3. Obsidian vault for things I need to read and edit by hand, written in Markdown: strategy, workflows, daily logs.

Each time an agent starts a work session, the agent reads the shared status file first, then does the work. When done, the agent records what was done. This way the next agent knows where the previous one left off. For example, an agent opening a session this morning will immediately see unfinished work from yesterday.

What is the website and what lessons did I learn?

This website is part of the system: the page you are reading, the course page and shop are all built and run by me with AI agents. Posts, tool library and prompts are managed in the admin panel. Agents read and suggest content through MCP, but everything agents write goes to a review queue before going live.

I drew three lessons:

• Start with one task, not the whole system. This system grew gradually from very small workflows.
• Write everything down. Status files and daily logs keep multiple agents working in sync with each other.
• Humans review the final step. AI is fast, but posting, sending or deleting still goes through my hands.

If you are also working alone and want to start automating, the next post I write is about how to pick the first task to hand over to AI.

Read more: I let AI agents build and run the whole website; The "SOP-ify" entire company process; "One week working 20 hours".