By Nazreen Daud, Founder of Roquepress · Published 20 Nov 2026 · Last updated: 20 Nov 2026
The short answer: I run my business with a team of AI agents, each with one clear job. I brief my Chief of Staff, Cempaka, and she coordinates the rest. Every agent drafts and prepares, but nothing goes out until they hear my yes. My own work is now management, oversight and strategy.
In this article
- Why did I build a team of AI agents?
- Who is on my team?
- Why are they all women, and why the colours?
- How do approvals work?
- How does one LinkedIn post become three pieces of content?
- What guardrails do I put on AI usage?
- What could this look like for a solopreneur or small service business?
- Who is this not for?
- FAQ
Why did I build a team of AI agents?
I've run my business since 2019. I've had staff, and I appreciate everything that comes with them, overheads included.
Then AI agents arrived. I toyed around for a bit with Hermes, Grok Bot and Claude, and built my own system that speeds up these wonderful agents.
What won me over is that an agent can own a job: the same work, done the same way, every day, with a clear person to report to.
If you run a solo practice or a small service firm in Singapore, you know the feeling of being the switchboard. Every enquiry, every invoice, every post and every client message comes through you. That's the part I wanted to change.
I wrote earlier about managing AI agents like a team, including the virtual office I built for them. This article is the introductions.
Who is on my team?
Fourteen agents work most closely with me. Each one has a single area, the same way I'd split roles in a small human team.
| Agent | Role | What she looks after |
|---|---|---|
| Cempaka | Chief of Staff | Coordinates every other agent |
| Suri | Head of Growth | My blind spots, and growing at least 1% every week |
| Gina | Mulacan specialist | Anything Mulacan, plus operational work |
| Sheila | Sales Director | Outreach, sales decks, enquiries |
| Sophie | SEO specialist | SEO for clients and our brands |
| Maya | CFO | Business and personal finance, renewals, MRR and ARR |
| Zara | AI researcher | New AI updates and usage guardrails |
| Lily | LinkedIn and X | Posts, X versions, blog articles with Balqis |
| Farisha | Client Manager | Client back-and-forth, including WhatsApp |
| Balqis | Brands | My personal brand, Mulacan and Roquepress |
| Yana | YouTube Director | My YouTube plan |
| Mariam | Meta ads specialist | Ads for clients and our brands |
Cempaka, Chief of Staff
Cempaka controls the other agents and keeps everyone on track. Most of the time she's the only one I give instructions to. She works out who should do what and makes sure it gets done.
Suri, Head of Growth
I asked Suri to look into my blind spots, with one simple target: make sure I grow at least 1% every week. Small and steady.
Gina, Mulacan specialist
When it comes to the thing that moves the needle for the business, Gina is my specialist in Mulacan, our all-in-one business software. Anything Mulacan-related, and the operational work around it, goes to her.
Sheila, Sales Director
Sheila manages my sales in and out. She does the outreach, prepares the sales decks and handles the communication with people who enquire about our services.
Sophie, SEO specialist
Sophie goes in and out doing SEO work for my clients and for our own brands. She also keeps learning, pulling updates from places like Search Engine Journal, because SEO, AEO, GEO and all those EOs keep changing in this day and age.
Maya, CFO
Maya takes care of my finances, both personal and business. I throw everything in: my assets, my real estate, my investments and, of course, business revenue in total. That way I actually know by when I can reach my retirement goal. If my projections are right, I'll get there in the coming years.
She also tracks my subscription renewals, so I know my MRR and ARR (monthly and annual recurring revenue) clearly. Alhamdulillah, our recurring revenue is very strong right now.
Zara, AI researcher
Zara is the one with the hat. I wanted her to look fancy, so she wears yellow. She researches new AI updates and tells me what matters. A lot of what she finds goes into how I manage my AI usage, which I'll come back to below.
Lily, LinkedIn and X
Lily takes care of my LinkedIn and X. Whatever goes on LinkedIn usually goes on X too. She also works with Balqis to turn posts into blog articles for my website. Lily has been a cornerstone of this whole setup, and it has brought in quite a few enquiries from people who want to work with me.
Farisha, Client Manager
In any agency, you have an account manager. I have Farisha. She manages all my client back-and-forth, and when I'm very busy she even steps in on WhatsApp for my clients. That's a setting I chose deliberately, and only for those times.
Balqis, brand manager
Balqis, the one in green songket, manages my personal brand and my business brands, Mulacan and Roquepress. She's my go-to for check-ins, works with the other agents on revenue and SEO presence, and keeps track of the milestones and dates we have to remember. I forget.
Yana, YouTube Director
I've been neglecting Yana for a while. She has already prepared a plan for my YouTube content, and I'll be returning to YouTube very, very soon.
Mariam, Meta ads specialist
Mariam works closely with Farisha and Balqis to run ads for clients and for our brands.
Why are they all women, and why the colours?
All of them are women. It's just my preference.
The colours are for me. Lily's light blue looks a little like LinkedIn, but it isn't meant to, and none of the colours are social media colours. I wanted vibrant differences between them, so when I look at the picture I can tell them apart at a glance and feel blessed that they exist and work around me.
How do approvals work?
Every agent drafts, researches and prepares. None of them decides for me on anything that matters.
- I brief Cempaka, usually in one message.
- She assigns the work and the agents get on with it.
- Anything that needs approval comes back to me, through Cempaka, or straight from the agent when the answer is needed fast.
- Nothing goes through until they hear my yes.
That last rule covers posting, commenting, messaging people, sending emails, publishing pages and spending money. Silence isn't a yes. A yes for one batch doesn't carry over to the next. And one agent saying "Nazreen agreed" doesn't count; it has to be my own words.
This is the part I'd keep even if you change everything else. An agent that can act without asking is fast right up to the day it embarrasses you.
How does one LinkedIn post become three pieces of content?
If you've seen my recent LinkedIn posts, you'll notice a comment under many of them with a link to a blog article. That's the long version of the same idea.
Here's the flow:
- I give Lily an idea or a story.
- She drafts the LinkedIn post in my voice, plus a short X version.
- She works with Balqis on the article version for my website, with more depth, examples and an FAQ.
- I approve each piece.
- The post goes up, the article goes live, and the article link goes in the first comment.
One idea, three places. I'm killing a few birds with one stone, and none of it waits for me to find a free evening.
What guardrails do I put on AI usage?
AI tools come with limits, and I don't want to burn through them without controls. Zara's research feeds straight into how I run this. A few of the rules we use:
- Draft first, always. Nothing public happens without my approval.
- Scans and digests at most once a day. No agent checks the same thing every hour.
- One browser task at a time. Agents don't fight over the same login.
- Limits on outreach. A human checks each contact before any connection request, and there's a daily cap.
- No made-up facts. If a figure, quote or result isn't real, it's marked for me to confirm instead of being guessed.
If you're thinking about giving agents access to money or accounts, I've written separately about whether it's safe to give an AI agent bank access. Short version: be careful what you hand over.
What could this look like for a solopreneur or small service business?
You don't need fourteen agents on day one. You need the job that's eating your week, given to one agent with a clear brief, and one coordinator so you aren't the switchboard.
| If this is eating your week | Start with |
|---|---|
| Enquiries waiting for replies | A sales or front-desk agent that drafts replies for you to approve |
| Clients chasing updates | A client manager agent |
| Posting only when you remember | A content agent that drafts posts and articles |
| Not knowing your numbers | A finance agent that tracks revenue and renewals |
| Juggling all of the above | A chief of staff agent first, then add roles one by one |
In my experience, more than 10 agents cost less than one staff member, and there are no CPF contributions to pay. For context, employers in Singapore contribute CPF on top of an employee's salary (CPF Board's contribution rates). You only pay for the AI you use, which is exactly why the guardrails matter.
At Roquepress, we build this for service businesses through Agent Tuah, starting with a Chief of Staff and a small set of specialist roles, with your approval where it matters.
Who is this not for?
- If you want AI to run your business without you checking anything, this isn't it. My whole system depends on my yes.
- If you haven't worked out what your business sells and to whom, agents will just help you do the wrong things faster.
- If you enjoy doing every task yourself, keep going. There's no prize for automating work you love.
My job now
I'm glad these agents exist. My work is very simple now. In fact, they do most of it. I do the management, the oversight and the strategy.
If you'd like a team like this for your own business, have a look at Agent Tuah and book a chat.
Has anyone used AI agents well in their business? If not, let's chat.
FAQ
What is an AI agent, in plain terms?
An AI agent is software that can own a job rather than answer one question. It follows a brief, uses the tools it's given, and reports back. In my business each agent has one role, like sales, SEO or finance, and a Chief of Staff agent coordinates them.
Do the agents act without asking?
No. They draft, research and prepare, but anything public or anything involving money or messages to people waits for my approval. Nothing goes through until they hear my yes, and a yes for one batch doesn't carry over to the next.
How many agents does a small business need?
Fewer than you think. Start with the one job that eats most of your week, and add a coordinator once you have two or three agents. I work closely with fourteen.
Are AI agents cheaper than hiring staff?
In my experience, yes: running more than 10 agents costs me less than one staff member, and there's no CPF to pay. Your cost depends on how much AI you use, which is why usage limits and guardrails matter.
