The hardest thing for the MD of a growing business is getting their week back. This is how one MD used AI to take the routine thinking and admin off his desk.
I built an AI leadership team in Claude this week: six named agents, each with one job, sitting alongside the real people who run the business with me.
I built this in Claude, the AI assistant we use most day to day. We’ll set you up with whichever AI tool suits your business.
How an AI Leadership Team Works
Every MD has a pile of jobs that never quite reach the top. Mine were the inbox, the pipeline, the website and the numbers I meant to check on Friday.
So I gave each pile an owner. Each agent is a Claude skill, which is a folder of written instructions Claude loads when the job comes up. Anthropic’s guide to what Claude skills are explains the mechanics. There’s no coding. The work is putting your standards on paper.
Here’s the team:
| Agent | Role | What it does for me |
|---|---|---|
| Bob | Chief of Staff | Takes every new task, scores it and hands it to the right owner |
| Quinny | Finance Director | Margin, cash, costs and what drives the value of the business |
| Wolf | Sales Director | Reads the pipeline and activity sheets, finds leads, keeps the CRM |
| Megan | PA | Turns Outlook, Teams, Slack and my calendar into one list for the day |
| Kyle | Marketing Manager | Sweeps the website for broken links and conversion leaks |
| Claire | Wellbeing | Checks the tone of a message before I send it |
Yes, they have names. It sounds daft until you try it. “Ask Wolf” is quicker to say than “analyse the sales pipeline against target”, and the team remembers who does what.
If you’d like help putting your own leadership jobs on paper, that’s exactly what we work through in a 30-minute call. Most owners have three of these agents’ worth of work sitting in their inbox already.
Start With a Chief of Staff, Not a Specialist
The most useful agent isn’t the clever one. It’s the one that decides what to do first.
Bob scores each incoming job from 1 to 5 on four things:
Money: the pound impact on sales, margin or cost
Urgency: a real deadline, or someone waiting
Only me: does it need the MD, or could someone else do it?
Effort: quick jobs score higher
A total of 15 or more means now. 11 to 14 means this week. 6 or under gets parked, and Bob says so plainly. Parking is a decision, not a failure.
That scoring is the bit I’d copy first. It works on a sticky note, never mind in an AI leadership team. It grew out of the day-to-day jobs I wrote about in what I actually use AI for.
Give Every Agent One Job and Hard Rules
Vague agents give vague answers. Each skill file says what the agent does, which files it reads, what the output looks like and what it must never do.
The rules matter more than the instructions. Mine are short:
Nothing gets sent, deleted, bought or changed without my yes
Rep comparisons and coaching notes are for my eyes only
Every number traces back to a file and a cell
Estimates are marked as estimates
If you’re handing an AI tool access to your inbox and files, the ICO’s guidance on AI and data protection is worth an hour of your time. Write your rules before you connect anything. If you want a second pair of eyes on them, that’s what advisory is for.
Put Your AI Agents on a Schedule
An agent you have to remember to ask is just another job on the list.
So four of mine run on their own. Megan sends a daily brief. Wolf does a Monday pipeline review. Kyle runs a site sweep. Claire does a Friday check-in on my week. Claude’s scheduled tasks run even when my laptop is shut, which is most of Sunday.
This is where AI agents for small business stop being a toy. The Monday review lands before the sales meeting, not after it.
What Didn’t Work First Time
I’d love to tell you it all clicked. It didn’t.
Megan couldn’t see Microsoft To Do through the connector, so flagged emails became her task list instead. Wolf couldn’t read two of the sales team’s diaries until each person shared them at the “Can view all details” level. I’d assumed that being able to see them in Outlook was enough. It wasn’t.
I also missed an obvious hire. I built five agents, looked at the list and realised there was no Sales Director. A glaring omission for a business with a field sales team. Wolf was added soon after.
And none of it replaces judgement. Quinny can tell you margin has slipped. It won’t tell you whether to have the difficult conversation with a supplier. That’s still your job.
Andrew’s Final Thoughts
An AI leadership team won’t run your business. It will stop the important-but-not-urgent jobs falling off the end of every week. Start with one chief of staff and a scoring rule. Add specialists only when you can write down exactly what good looks like. If you’d rather not start from a blank page, our workshops build the first two agents with you.
AI Leadership Team FAQs
What is an AI leadership team?
An AI leadership team is a set of AI agents, each set up to do one leadership job, such as finance, sales or diary management. In Claude, each one is set up as a skill: written instructions, the sources it reads and the rules it must follow. They support the people running the business. They don’t replace them.
Do I need to code to build AI agents?
No. Skills are written in plain English: what the job is, where the information lives, what the result should look like and what’s off limits. The hard part is being clear about the standard you expect. Few owners have ever written that down.
Is it safe to connect AI agents to email and company files?
It can be, if you set the rules first. Give agents read access where you can, and require your approval before anything is sent, deleted or bought. Keep sensitive comparisons private, and check your data protection position before connecting customer data.
How long does it take to build an AI leadership team?
A first working version of each agent takes an hour or two once you know what you want. The time goes on fixing access, such as calendar sharing and file locations, and on tightening instructions after the first few runs. Begin with a single agent and add the next once it earns its place.
Which AI agent should a small business build first?
Build the triage agent first: one that sorts and prioritises incoming work. It saves time on day one and makes every other agent more useful, because it decides what each of them should pick up. Next, build whichever job you most often put off. For many MDs, that’s the weekly sales report.
We’re not IT people. We’re business leaders who’ve put AI to work across sales, finance, operations and marketing. We can set this up for you, train your team and hand it over, or keep running it for you each month.
Growing pains like this are normal in a business that has grown quickly, and they’re fixable. Take the free Growing Pains Check to see where your business stands in three minutes, or book a free call about a Growing Pains Diagnostic: half a day with you and your business, then a written 90-day plan with the three changes that matter most.
Leave a Reply