# Solo business with AI: a map of 7 departments and what to delegate to AI

URL: https://opcaios.org/en/blog/ai-giup-mot-nguoi-van-hanh-doanh-nghiep
Date: 2026-11-05
Author: Lê Quốc Nghĩa

# One-Person Business with AI - What Does It Mean?

One-person business with AI means the business owner holds the decision-making role, sales and responsibility, while repetitive tasks across other departments are drafted, handled or checked by AI agents and automation tools. This is feasible because language models are good at writing, summarizing and classifying (see [What is LLM?](/blog/llm-la-gi)), but not every task should be delegated: tasks that cannot be undone, carry legal risk or need personal credibility still require human review. For example, when using Dify as an assistant on a website and [n8n](/tien-ich/n8n) for automation flows, one person still retains final review rights for all outgoing content; the information in this post was compiled on 24/10/2026.

![Diagram of seven one-person business departments: sales, marketing, customer care, operations, finance, research, systems; each department has tasks delegated to AI and tasks kept by the person](https://opcaios.org/images/blog/ai-giup-mot-nguoi-van-hanh-doanh-nghiep/so-do.webp)
*Delegate repetitive work to AI, keep decisions and risk points. Diagram: OPCAIOS.*

## Delegation principle: delegate drafts, keep decisions

The "delegate drafts, keep decisions" principle means AI creates drafts and handles repetitive work, while the business owner decides and takes responsibility. Hypothetical example: AI drafts a quote for a client, the business owner reviews the numbers and terms before sending. Use three questions to classify a task, in order:

1. **Is it repetitive and does it have a clear pattern?** If yes, automate it.
2. **What are the consequences of getting it wrong?** If it cannot be undone (sending to clients, payment, public posting), it needs human review.
3. **Does it need credibility or personal relationships?** If yes, keep it for the person.

## Map of seven departments

| Department | Delegate to AI | Keep for person | Risk to control |
|---|---|---|---|
| **Sales** | Draft outreach, summarize calls, update CRM, follow-up reminders | Deal closure conversations, price negotiation, relationships | Overpromising, sending to wrong client |
| **Marketing** | Brainstorm ideas, write posts, convert formats, schedule posts | Positioning, brand voice, publish approval | Generic content, wrong numbers |
| **Customer care** | Answer FAQs, classify requests, draft responses | Complaints, exceptions, refunds | Wrong policy answers |
| **Operations, delivery** | Checklists, records, progress reports, task reminders | Quality control of deliverables | Missing client requirements |
| **Finance** | Classify transactions, invoice reminders, compile reports | Tax numbers, spending decisions, reconciliation | Accounting errors |
| **Research** | Find sources, summarize, compare competitors | Strategic conclusions | Unverified sources |
| **Systems, technology** | Write code, test, monitor, alerts | Access rights, security, backups | Security flaws, data loss |

## Three layers to organize: context, process, control

Three layers of organization is how to arrange things so one person can coordinate multiple AI agents without losing control. First is context, second is process, third is control. Hypothetical example: the business owner writes quote rules into a note, builds a process using n8n or Dify, and sets alerts via Telegram. Details:

- **Context:** one place to store information about the business, customers, tone, rules; AI reads from here instead of guessing (see [What is Second Brain?](/blog/second-brain-la-gi) and [What is RAG?](/blog/rag-la-gi)).
- **Process:** repetitive steps built with automation tools like [n8n](/blog/n8n-la-gi) or platforms like [Dify](/blog/dify-la-gi), with stopping points for human review.
- **Control:** logs, alerts, minimum permissions, backups; see [What is MCP?](/blog/mcp-la-gi) for the principle of granting minimum permissions to agents.

## What does a sample workday of a one-person business owner look like?

A sample workday breaks into four time slots, where AI handles summarizing and drafting while the business owner handles review and decisions. Hypothetical example: mornings take only about 10 minutes to read the summary.

| Time | AI's work | Your work |
|---|---|---|
| Morning | Summarize new emails, new leads, system alerts into one brief | Read 10 minutes, pick three priorities |
| Midday | Draft post copy, customer responses, quotes | Review, edit, send |
| Afternoon | Update CRM, run reports, check systems | Call clients, meetings, close deals |
| End of day | Log entries, prepare tomorrow's summary | Review, decide adjustments |

This is an example framework, not a measurement; actual time depends on your industry and volume.

## Things You Should Not Hand Over Completely to AI

- Sending quotes, contracts, and invoices to customers without reviewing them first.
- Transferring money, issuing refunds, or changing payment terms.
- Publishing content publicly in your name without approval.
- Processing sensitive personal data without security rules in place.
- Making hiring, firing, signing, or legal compliance decisions.

## OPCAIOS's Experience

We apply the principle of "hand over drafts, keep decisions" correctly. Articles, English translations, and draft emails written by AI go through an approval queue and don't publish themselves; the website assistant (built with Dify) transfers contact information into the CRM; a monitoring system checks services regularly and sends alerts via Telegram; and a shared status file serves as the source of truth for agents. The work we still keep for people is talking with customers, deciding priorities, and approving everything public. We haven't measured time saved by department yet, so we don't include a ratio in this post.

Read more: [How to "SOP-ify" your entire company](/blog/sop-hoa-cong-ty-tu-chay-khi-di-du-lich); [How to create a Voice AI Agent with Vapi.ai to automatically call and confirm orders](/blog/voice-ai-agent-vapi-xac-nhan-don-hang).

## Frequently Asked Questions

### Can one person really run an entire business with AI?

Yes, at a small scale, if your product is simple, processes are clear and repeatable, and you have someone reviewing decisions at risk points. OPCAIOS doesn't have overall data on the success rate of one-person businesses, so this is a practical observation, not a research finding. The hardest part is still sales and building trust, because it needs credibility and personal relationships.

### Which department should I automate first?

Start with work that takes a lot of time, repeats often, and causes little damage if done wrong—usually drafting, summarizing, and reminders. First, list weekly repeating tasks and apply the three sorting questions above. Next, pick one task that's easy to fix if wrong and set a stopping point for review. Finally, expand to hard-to-undo tasks like quotes or payments. Marketing and research are usually easier to start with than finance or systems.

### Can AI replace an employee?

AI can replace some repeating tasks, not entire roles. Drafting greetings, answering common questions, or writing code can be handed to AI as drafts; complaints, refunds, hiring, and signing still need people. OPCAIOS hasn't measured time saved by department, so we don't include staffing replacement ratios in this post. See the cost calculation in [the post about AI tool costs for OPC](/blog/chi-phi-cong-cu-ai-thay-3-nhan-su).

### What tools do I need to start?

You need four parts: one AI assistant, one automation tool like n8n or Dify, a place to store context, and a simple CRM. For example, you can start with just an AI assistant and a customer spreadsheet, then add automation later. First, pick repeating work; next, write down the process; finally, connect the tools and set a stopping point for review. OPCAIOS doesn't recommend buying everything at once, because each added tool is a cost and a potential failure point; see [AI tool set for the OPC model](/blog/bo-cong-cu-ai-cho-mo-hinh-opc). Learn how to build this system in [OPCAIOS's AI application training course](/services/dao-tao-ung-dung-ai).

*Source: the work-division map is OPCAIOS's consulting framework based on systems-building experience, not research data; knowledge of language models and MCP comes from OPCAIOS's related posts (read on 24/10/2026). The experience is OPCAIOS's.*
