Short answer
A one-person company does not need ten disconnected AI tools. Begin with three workflows: research and decision support, customer follow-up, and delivery and quality review. Define the role, allowed inputs, output standard, escalation conditions and human owner for each. AI can take on team-like work, but must not independently commit, pay, contract or make high-impact decisions.
A collection of tools is not a team
I have met founders who subscribe to one tool for writing, another for images, another for meetings and another for research. The screen looks advanced; the work remains exhausting. The tools have no roles or handovers. The owner must paste the context again, explain the standard again and carry each result to the next place.
A real AI team should resemble a small company. Each role knows what it receives, what it must deliver, when it must ask a person and where completed work goes. Once those rules are retained, one person’s judgement can begin to become the company’s method.
Workflow one: research and decision support
This workflow does not let AI decide for the owner. It keeps the owner from becoming lost in information.
Inputs may include customer questions, public competitor material, previous proposals, sales outcomes and current constraints. The AI role organises sources, marks contradictions, lists unknowns and proposes testable options. The completion standard requires a source for each conclusion, explicit assumptions and a next step that can be tested within a bounded time.
The person chooses the question, judges risk and approves action. AI must not present unsourced market claims as facts or make a major investment decision merely because it processed many documents.
Workflow two: customer follow-up
A one-person business often loses small promises rather than large projects: reply by Friday, send one document, reconnect in three weeks. They sit across email, WhatsApp, meeting notes and memory until a busy day buries them.
The follow-up assistant can extract needs, commitments, next action, owner and date after each interaction; present a daily human action list; and prepare drafts. Sending needs a risk boundary. Routine confirmations may go after review, while quotations, legal commitments, discounts and sensitive content require human approval.
Success is not an empty inbox. It is a customer who understands the next step and a company that can trace each promise to its conversation.
Workflow three: delivery and quality
When the owner sells and delivers, a promise is vivid at the time of sale and vague at completion. A delivery assistant turns approved scope into a checklist, timeline and acceptance points; compares work against the original request before submission; and preserves feedback and corrections.
AI can perform a first review, not approve itself. Numbers, legal, health, finance, brand claims and client-confidential material need the right human check. Version history also matters: the “latest” file must not be a newly generated version that has drifted from approved content.
Hypothetical example: a consultant stops restarting each day
This is a teaching scenario, not a Smark Global client case.
A Hong Kong independent consultant researches markets, answers enquiries, holds meetings, writes proposals and delivers reports each week. AI is already in use, but the company introduction and writing requirements are pasted afresh every time. Tone changes and client constraints occasionally disappear.
The consultant creates three small specifications. Research must include sources and unknowns. A customer summary must include commitments, dates and next action. Delivery review must check each item against approved scope. All three workflows share one approved description of services, voice, pricing principles and matters that can never be automated.
After several weeks, the main change is not a promise of ten times more work. It is no longer having to remember from zero how the company should work. A future assistant can also learn without relying only on oral explanation. The one-person business has begun to acquire organisational capacity.
Permissions, cost and stopping boundaries
An AI team needs permissions like a human team. A research assistant may read public sources, not every client file. A follow-up assistant may draft, not necessarily send. A delivery assistant may check documents, not approve payment or sign contracts.
Each workflow needs a cost ceiling, error log and stop control. Missing inputs, low confidence, sensitive content or work outside scope should trigger human review. Mature automation is not automation that never asks. It knows when it must ask.
AI team role card
- For every workflow, define:
- role name and purpose;
- information it may access;
- required output format;
- completion and quality standard;
- actions it must never take alone;
- escalation conditions;
- final human owner;
- cost, error and version records.
Further reading
Want the full version of this approach?
If you want one person’s experience to become a company capability, view Harry Lui’s book on Amazon or register for preview material and enterprise AI notes.
Harry Lui's book works through the same path in full: from business context to company data to a reusable operating system.
Source: Smark Global editorial, paraphrasing ideas from Harry Lui's book
Last updated: 2026-09-13（first published: 2026-09-13）
