Keeping a team of AI agents in check across several small businesses
I built a control hub for the AI agents that do routine work across several small businesses, recording which bot serves which business, who may talk to it, the rules it follows and everything it said and did.
- Sector
- AI operations for small businesses (cross-sector)
- Stack
- Next.js, Postgres, Azure, MCP, REST API
- Status
- Live, in active testing
The problem
Once a business has more than a couple of AI assistants, basic questions get hard to answer: which bot handles which company, who is allowed to instruct it, what it sent yesterday and why, and whether the old prompt can go back when a change breaks something.
Without answers, AI becomes a risk rather than a saving.
What I did
I designed and built a hub that sits beside the bots rather than in their way. It holds the structure (accounts, the companies inside them, departments and bots), the contacts and roles allowed to reach each bot on each channel, and rules that apply from the top down so a lower level can never override a higher one.
Every bot logs what it did to one central record, prompts are backed up by version, and problems are filed once and remembered, so the same issue isn't raised twice. Any AI tool can connect to it through a standard protocol (MCP) or a REST API with its own key.
Email and WhatsApp traffic comes in through the shared inbox I built for small firms.
Tools and stack, in plain words
A Next.js app with Postgres and Azure file storage, an MCP server so AI assistants can connect directly, per-bot access keys, and automated tests for routing, permissions and every signed-in screen. It deploys to Azure automatically.
Outcome
Live, with ongoing testing. The AI agents across a group of small businesses now have one place that says what each one is for, who it answers to and what it has done, with a full record behind it.
What it shows about how I work
This is the one-person-plus-AI model in practice. I let AI take on real work, and I build the controls that make that safe to do.
FAQ
How do you keep control of several AI agents in a small business?
Him Yip's approach is a central registry: each bot is tied to the businesses it serves, only named contacts can instruct it, rules apply from the top down, and every action and prompt change is logged.
Can AI agents from different providers work with one control system?
Yes. The hub is provider-neutral; any assistant that supports MCP or a REST API can connect with its own key.