我如何为一人公司设计 AI 系统（以及我拒绝自动化的部分）

I founded Write Right in 2016 and grew it to 75 people. I exited that company in 2023. It was a real operation: people, process, clients, an org chart that existed for a reason. Before that I had already exited Dad of Ad, in 2021. In October 2026 I exited Blushush. All three were bootstrapped.

After the third exit I went back to building alone. The instinct, if you have ever run a company of that size, is to rebuild the machine: hire the functions, recreate the layers, put a person or a tool under every box on the old chart. That is the wrong instinct for a one person business.

A one person business is not a 75-person company with 74 empty chairs. It is a different design. It has the same kinds of work — research, delivery, follow-up, a public record of who you are — but it cannot afford the same structure. The work has to travel a short path from a decision to a finished thing. If you copy the headcount logic, you will spend your week managing a system that was built to manage other people.

I am a serial entrepreneur and personal branding consultant in Ahmedabad, co-founder and CEO of Ohh My Brand, and I am now building Solo Not Small, an AI initiative that helps solopreneurs and small business owners scale with AI. I started as a mechanical engineer at AMW Motors in Bhuj. The shop floor and the 75-person company taught me the same lesson: copy the work, not the org chart.

Map the jobs before you pick a tool

People ask which model or agent they should adopt. That is usually the wrong first question. Start by writing down the work you already do in a week, in the order it actually happens.

When I do this for a one person business, the list almost always falls into four groups.

The first is revenue work: the call, the proposal, the follow-up, the brief you write before you speak to anyone. The second is delivery: the thing you promised, in the form you promised, at the standard you are willing to put your name on. The third is the record: the about page, the profiles, the short answer you give when someone asks what you do. The fourth is learning: the notes you take after a call, the decision you do not want to make twice, the research you will need again next month.

A 75-person company spreads those groups across teams. A one person business holds all four, and the leak is usually in the joints: thin notes after a call, a public record that still describes last year’s company, a late follow-up because the brief was never written down. AI is useful at those joints. It is much less useful as a costume for a company you do not have.

The first job I systemise is never the public one

The temptation, once you have a capable model, is to start with the visible output: a post, a landing page, a thread that makes the one person business look busier than it is. I think that is backwards.

The first job I systemise is the one that happens before anyone sees me: the briefing before a sales call, the research that stops me improvising, the notes after a conversation, the checklist I have already run a hundred times and still sometimes skip. Those jobs are repetitive, they have a clear definition of done, and they do not require the model to invent a personality.

This is also where a one person agency gets into trouble. An agency is a special case of a one person business: the founder is the delivery team and the brand. If you automate the public surface first, you can end up with more content than judgement, and the two no longer match.

A practical test I use: if I cannot explain the job in one sentence, and point to an example of it done well, I do not hand it to a model yet. I do the job myself one more time, write the brief from that pass, and only then ask the system to draft, extract, or check. The model is staff for work I already understand, not a voice.

Treat AI like staff, not like a personality

Staff need a brief. They need the facts they are allowed to use. They need to know what they must not invent. A one person business that skips this will get fluent answers that are slightly wrong — a date moved, a role inflated, a product described as something it is not.

I am strict about the last point in my own work. Solo Not Small is an AI initiative for solopreneurs and small business owners. It is not a course. My exits have names and years. I do not ask a model to decorate those facts, and I do not let it create metrics I have not supplied.

That sounds obvious. It is not how most people brief a tool. They paste a loose prompt, get a confident paragraph, and publish it because it sounds like a company. For a one person business, that is a costly habit. There is no editor down the hall. The person who notices the error is often the person you were hoping to work with.

So the system I actually want is boring. A short brief for each repeating job. A short list of facts the model may use. A human pass on anything that will be seen in public. A place to put the output so I can find it again. That is an operating system. A pile of chat windows is not.

What I refuse to automate

There are jobs I will not hand over, even when the draft would be fine.

I will not let a model decide what the company is. The public record — the about page, the profiles, the sentence I use in a room — has to be finished by me. If those surfaces disagree, I do not have a marketing problem. I have an operations problem. I work on that question as a personal branding consultant, and I have written about related themes for Forbes India and Entrepreneur India, and as a member of the Fast Company Executive Board. Consistency is a system, not a slogan.

I will not let a model invent a client, a number, or a result so the work sounds bigger. And I will not automate the last judgement on a piece of work that goes out under my name. A draft can come from a system. The standard cannot. Write Right needed layers so quality could survive when I was not in the room. A one person business does not have that excuse, and it does not need that overhead.

Volume is easy now. The scarce thing is a short path from a real decision to a finished piece of work that is still true.

A short way to start this week

If you run a one person business and you want a system rather than another tool, I would do this, in order.

Write the four groups of work as they really happen in your week. Pick one pre-public job — a brief, a follow-up, a research pack — and describe it in one sentence plus an example. Run that job with a model only after the brief exists. Keep a single place for the output. Then look at your public record and check whether it still matches the company you are actually running. If it does not, fix the record before you generate more of it.

The same sequence helps a one person agency, where the pressure to look like a 20-person shop is stronger. You do not need to perform a firm you do not have. You need the joints to work, and you need the record to stay true.

That is the work I am doing myself, and it is the problem Solo Not Small is built around. I have been the engineer in Bhuj and the founder of a 75-person company. I am now back at one desk. The useful inheritance is standards, not structure. The rest is judgement, and I would like to keep that part human.

Serial entrepreneur and personal branding consultant based in Ahmedabad, India. Co-founder and CEO of Ohh My Brand, now building Solo Not Small, an AI initiative helping solopreneurs and small business owners scale with AI. More at bhaviksarkhedi.com.

Find me on Hashnode, HackerNoon, Quora and Strikingly.