
My solo AI operations system through October 2026
A solo AI operations system is the set of tools and agents I've built to handle repetitive work instead of hiring people. I work alone with no assistant or operations team, but each day still has dozens of small tasks: note-taking, answering messages, saving documents, monitoring servers, writing content, downloading and processing videos. This post describes how that system is running through October 2026, using a Mac, a Windows machine, n8n, Obsidian and Telegram, so you can see how one person alone can organize AI. This is my own approach, not yet a general recommendation. 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-divided system work like a small team?
My system is a small team of three "members," each doing one clear job instead of using one chatbot for everything. The way I run it through October 2026, the three parts are:
- Command brain running on a Mac: where data lives, where automated processes run and where the gateway to AI models sits.
- Executive arm running on a Windows machine: an AI agent that takes work 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 the agent can read.
The two machines connect via a private network, so the agent on Windows 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 receives the task, 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 that I've installed myself. Each workflow is a chain of steps: receive data, process it, call AI if needed, then record the result in the right place. Agents also send tasks to each other via a webhook on n8n. Each message has the same four-part template: who sends it, who it goes to, what type of work (assign task, return result, alert, status update) and content. For example, one workflow receives a message via Telegram, calls a model through 9Router, then records the result in Obsidian. Thanks to this common template, 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 powerful model. Next, simple work uses a cheap model. This way I'm not locked to 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:
- Short-term memory for events, habits, often-used parameters. Agents record and read it back quickly.
- Knowledge database (Postgres with semantic search) for long documents, scripts, large knowledge. Documents are split into chunks and turned into vectors so AI finds the right chunk.
- Obsidian vault for what I need to read and edit by hand, written in Markdown: strategy, work processes, daily logs.
Whenever an agent starts a work session, the agent reads the shared status file first, then gets to work. When done, the agent records what it did. 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're reading, the course page and shop are all built and run by me with the AI agent. Posts, tool libraries and prompts are managed in the admin panel. The agent reads and suggests content via MCP, but everything the agent writes goes into a review queue before going live.
I learned 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.
- People review the final step. AI works fast, but posting, sending or deleting still goes through my hands.
If you're also working alone and want to start automating, in my next post I'll write about how to pick the first task to hand over to AI.
Read more: I let AI agents build and run my entire website; The "SOP-ify" process for the entire company; "A 20-hour work week".
Translated from Vietnamese with AI. Read the original.
Want to apply this in your business?
Book a free 30-minute call (Vietnam time, in English or Vietnamese).
